Download - Teradata Factory - Online Tutorials ( Books & Videos)dbmanagement.info/Books/MIX/TF_Book1_Version_6.2.0_TeraData.pdf · Teradata Factory Student Manual #1 Teradata Concepts NCR System

Teradata Training

Teradata Factory

Student Manual #1

NCR Teradata TrainingCourse # 9038Revision 6.2.0

Course Overview Page 0-3

Module 0

NCR Proprietary and Confidential

Course Overview

Teradata Factory

Student Manual #1

Teradata ConceptsNCR System Architectures

Teradata Division Training

Page 0-4 Course Overview

Sixth EditionJuly, 2003

NCRIntellectual PropertyIP

Developed by:Teradata Training Technical Development and Delivery

Copyright 2002 and 2003 by NCRDayton, Ohio

All Rights ReservedPrinted in U.S.A.


Table of Contents

COURSE DESCRIPTION ...................................................................................................................................... 0-6TRADEMARKS....................................................................................................................................................... 0-6WHO SHOULD ATTEND...................................................................................................................................... 0-8PREREQUISITES ................................................................................................................................................... 0-8CLASS FORMAT.................................................................................................................................................. 0-10CLASSROOM RULES ......................................................................................................................................... 0-10OUTLINE OF MAJOR SECTIONS.................................................................................................................... 0-12TERADATA CERTIFICATION.......................................................................................................................... 0-14TERADATA CERTIFICATION – MATERIALS TO FOCUS ON ................................................................. 0-16ADDITIONAL RESOURCES .............................................................................................................................. 0-18



Course DescriptionThis course provide information on the following major topics:

• Teradata Concepts• 48xx/52xx and 49xx/53xx Architectures• Teradata Physical Database Design• Teradata SQL ANSI Differences for Version 2• Teradata Application Utilities• Teradata Database Administration

TrademarksThe following names are registered names or trademarks and are used throughout thismanual.

AT&T is a registered trademark of the American Telephone and Telegraph Company.BYNET is a trademark of NCR.i386, i486, Pentium and Pentium Pro are trademarks of the Intel Corporation.EMC2 is a registered trademark of the EMC2 Corporation.JobAcct, Copyright © 1988-1994 by UniSolutions Associates.Kerberos, copyright © 1989 by the Massachusetts Institute of Technology.Micro Channel is a trademark of IBM Corporation.Microsoft, MS-DOS, and Windows are trademarks of Microsoft Corporation.MOTIF is a registered trademark of Open Software Foundation, Inc.NCR is a registered trademark of the NCR Corporation.ORACLE is a registered trademark of the Oracle Corporation.ProComm is a registered trademark of DATASTORM Technologies, Inc.Siemens is a registered trademark of AKTIENGESELLSCHAFT.SQLBase, SQLWindows, and SQLTalk are registered trademarks of Gupta Technologies, Inc.SYBASE is a registered trademark of Sybase, Inc.Symbios, Symbios Logic, and RAID Manager are registered trademarks of LSI Logic, Inc.TERADATA is a registered trademark of NCR.TOP END is a registered trademark of BEA Systems, Inc.UNIX is a registered trademark of X/Open Company Limited.X/Open is a trademark of X/Open Company Limited.Ynet is a trademark of NCR.

In addition to these product names, all brand and product names in this manual aretrademarks of their respective holders.


Course Description

Description

The primary focus of this ten day course is to teach you about the design,implementation, and administration of the Teradata Database.

The major topics in this course include:

• Teradata RDBMS features and functions

• The parallelism of the Teradata RDBMS

• How Teradata is implemented on NCR systems (e.g., 4951/5351)

• How to perform physical database design for Teradata RDBMS

• Teradata SQL ANSI Differences for Version 2

• How to create and populate the Teradata RDBMS using the Teradataapplication utilities

• How to perform common administrative functions for the Teradata RDBMS



Who Should AttendThe Teradata Factory is a learning event for RDBMS experienced individuals who need tolearn the Teradata RDBMS. This course is designed for Teradata practitioners who need toget hands-on practice with the Teradata RDBMS in a learning environment.

• NCR Professional Services Consultants

• NCR Channel Partners

PrerequisitesAn understanding of relational databases, SQL, and the logical data model is necessarybefore attending this course.

Experience with NCR large systems, relational databases and SQL, and an understanding ofthe UNIX operating system is useful, but not required before attending this course.

NCR offers the following Web Based Training classes that provide information aboutTeradata concepts and SQL.

• Overview of Teradata

• Teradata SQL

– Beginning Teradata SQL Part 1– Beginning Teradata SQL Part 2– Teradata V2R2 Features and Functions– Teradata V2R3 Features and Functions– Teradata V2R4 Features and Functions– Teradata V2R5 Features and Functions


Who Should Attend and Prerequisites

Who Should Attend

This course is designed for ...

• NCR Teradata Professional Services Consultants

• NCR Channel Partners

Prerequisites

Required:

• An understanding of the logical data model, relational, SQL, and dataprocessing concepts.

Useful, but not required:

• Experience with relational databases and SQL

• Experience with UNIX SVR4 MP-RAS

• Experience with NCR large systems



Class FormatThis ten-day class will be conducted as a series of lectures with classroom discussions,review questions, and workshops.

Classroom RulesThe classroom rules are listed on the facing page.


Class Format and Rules

Class Format

This ten day class consists of ...

• Instructor presentations

• Class discussions

• Workshop exercises

Classroom Rules

The classroom rules are …

• Turn off your cellular phones and pagers.

• Leave your laptop closed in the classroom during lecture.

• Come to class on time in the morning and after breaks.

• Enjoy the two weeks.



Outline of the Four BooksAn outline of the four Teradata Factory books is described on the following page. Majortopic examples are listed for each book.


Outline of the Four Books

1. Teradata ConceptsTeradata features and functionsParallelism and Teradata

NCR System ArchitecturesCharacteristics of NCR 495x and 535x systems - typical configurationsDisk Array subsystems and how Teradata utilizes disk arrays

2. Teradata Physical Database DesignPrimary and secondary index selection; partitioned primary indexesHow the Teradata database works

Features of Teradata SQLSQL ANSI syntax & features; Teradata and ANSI transaction modesTemporary tables, system calendar, & V2R5 System Limits

3. Teradata Application UtilitiesLoad utilities (e.g., BTEQ, FastLoad, MultiLoad, and TPump)Export utilities (e.g., BTEQ and FastExport)

4. Teradata Database AdministrationDictionary tables and views; system hierarchy and space managementUsers, Databases, Access Rights, Roles, and Profiles



Teradata CertificationThe flow chart on the facing page illustrates the different levels of Teradata Certification.

The names of the actual Certification tests are:

1 – Teradata Basics2 – Teradata Physical Implementation3 – Teradata SQL4 – Teradata Administration5 – Teradata Design6 – Teradata Application Development

The Teradata Factory assumes that you have knowledge of the SQL syntax. To gain thisknowledge to pass the Teradata SQL Certification test, the following Web Based Trainingclasses in SQL are available.

- Beginning Teradata SQL Part 1- Beginning Teradata SQL Part 2- Teradata V2R2 Features and Functions- Teradata V2R3 Features and Functions- Teradata V2R4 Features and Functions- Teradata V2R5 Features and Functions

To obtain the knowledge required to pass the Teradata SQL Certification test, you need tocomplete the first 4 levels of the SQL WBT classes listed above.

Additional information can be located at this web site:

http://www.global.ncreducation.com/teradata

http://www.global.ncreducation.com/teradata


Teradata Certification

Certified TeradataSQL Specialist

3

Certified TeradataApplication Developer

6

Certified TeradataImplementation Specialist

2

Certified TeradataAdministrator

4

Certified Teradata Professional 1 (Basics)

Certified TeradataDesigner

5

This course helps prepare you for these Teradata Certification tests.



Teradata Certification – Materials to Focus onThe flow chart on the facing page contains the areas or topics that you need to focus on andstudy to prepare for the Teradata Certification tests.


Teradata Certification – Materials to Focus On

Certified Teradata SQL Specialist Test 3

Attend WBT or Customer Education SQL classes

Certified TeradataApplication Developer

Test 6

80% - TF All Books20% - Advanced SQL, OLAP,

Preprocessor, CLI, ODBC, DATE, etc.

Certified Teradata Implementation SpecialistTest 2

TF Book 1 modules 10 and 13,Book 2 ALL,

Book 3 modules 1 and 9,Book 4 modules 1 through 3

Certified TeradataAdministrator

Test 4

75% - TF Book 4 ALL25% - Scattered in TF

Books 2 and 3

Certified Teradata ProfessionalTest 1 (Basics)

Teradata Factory (TF) Book 1 modules 1-10Teradata Factory (TF) Book 3 module 1

Certified TeradataDesigner

Test 5

70% - TF Book 2 ALL30% - Scattered in TF

Books 1, 2 and 3

This course helps prepare you for these Teradata Certification tests.



Additional ResourcesThe facing page lists various WEB sites and resources that provide additional informationabout Teradata.

Electronic versions of materials can be located at the following WEB site:

http://www.ncru.ncr.com/ncru/institutes/td/coursrch.asp

Logon as John Public (default user name) and use the default password.

These materials are to only be used by PS consultants as reference materials.

These materials are not to be distributed to customers in either electronic or printed formats.They are also not to be used as teaching materials by associates outside the TeradataDivision Training organization without the express consent of the Teradata DivisionTraining organization.

An additional site is the Teradata Global Sales Support web site.

http://gss.sandiegoca.ncr.com/gsstech/

http://www.ncru.ncr.com/ncru/institutes/td/coursrch.asp

http://gss.sandiegoca.ncr.com/gsstech/


Additional Resources

Teradata Certification Programhttp://www.global.ncreducation.com/teradata

NCR Documentationhttp://www.info.ncr.com (Teradata Warehouse Solutions, search for Product ID B035)

NCR Discussion Lists (e.g., ATR)http://mlist.ncr.com/#subs

NCR Universityhttp://www.ncru.ncr.com

Teradata SQL WBT courseshttp://www.learnzone.ncr.com/learning/sql/index.htm

Teradata Configuration WBT Coursehttp://www.learnzone.ncr.com/learning/config5/index.htm

Teradata Warehouse Builder WBT Coursehttp://www.ncru.learnzone.ncr.com/learning/td/twb/

Teradata System Performance Informationhttp://infobahn.sandiegoca.ncr.com/gsstech/techinfo/td_perf/



Notes

Teradata Product Overview Page 1-1

Module 1


Teradata Product Overview

After completing this module, you will be able to:

• Describe the purpose of the Teradata product

• Give a brief history of the product

• List major architectural features of the product


Page 1-2 Teradata Product Overview

Notes


Table of Contents

WHAT IS TERADATA?......................................................................................................................................... 1-4TERADATA - A BRIEF HISTORY ...................................................................................................................... 1-6HOW LARGE IS A TRILLION............................................................................................................................. 1-8THE TERADATA CHARTER............................................................................................................................. 1-10TERADATA’S COMPETITIVE ADVANTAGES............................................................................................. 1-12REVIEW QUESTIONS......................................................................................................................................... 1-14



What is Teradata?Teradata is a Relational Database Management System (RDBMS) for the world’s largestcommercial databases. Current technology permits databases of up to 100 terabytes in size.This characteristic makes Teradata an obvious choice for large data warehousingapplications, however the Teradata system may also be as small as 10 gigabytes. With itsparallelism and scalability, Teradata allows you to start small with a single node and growlarge with many nodes through linear expandability.

Since 1993, Teradata has been available on UNIX-based platforms and as of 11/98, alsoavailable on Windows NT. In 2000, Teradata was also made available on the Windows2000 operating system. Teradata is an open system and is compliant with industry ANSIstandards.

Teradata is comparable to a large database server, with multiple client application makinginquiries against it concurrently.

The ability to manage terabytes of data is accomplished using the concept of parallelism,wherein many individual processors perform smaller tasks concurrently to accomplish anoperation against a huge repository of data. To date, only parallel architectures can handledatabases of this size.


What is Teradata?

Teradata is a Relational Database Management System (RDBMS).

Designed to run the world’s largest commercial databases.• Preferred solution for enterprise data warehousing• Executes on UNIX MP-RAS and Windows 2000 operating systems• Compliant with ANSI industry standards• Runs on a single or multiple nodes• Acts as a “database server” to client applications throughout the enterprise• Uses parallelism to manage “terabytes” of data

TeradataDATABASE

Win XPWin 2000

IBMMainframe

UNIX



Teradata - A Brief HistoryThe Teradata Corporation was founded in 1979 in Los Angeles, California. The corporategoal was the creation of a ‘database computer’ which could handle billions of rows of data,up to and beyond a terabyte of data storage. It took five years of development before aproduct was shipped to a first customer in 1984. In 1982, the YNET technology waspatented as the enabling technology for the parallelism that was at the heart of thearchitecture. The YNET was the interconnect which allowed hundreds of individualprocessors to share the same bandwidth.

In 1987, Teradata went public with its first stock offering. In 1988, Teradata partnered withthe NCR Corporation to build the next generation of database computers. Project 90developed the advancements that made possible the 3600 system. Before either companycould market its next generation product, NCR was purchased by AT&T corporation.Teradata was also purchased and folded into the NCR structure. The new company wasnamed AT&T GIS.

In 1996, AT&T spun off three separate companies, one of which was NCR which thenreturned to its old name. By 1997, NCR had become the world leader in scalable datawarehouse solutions, largely due to the strength of the Teradata database.


A Brief History1979 - Teradata Corp founded in L.A. Cal

- Development begins on a massively parallel computer

1982 - YNET technology is patented

1984 - Teradata markets the first database computer DBC/1012- First system purchased by Wells Fargo Bank of Cal.- Total revenue for year - $3 million

1987 - First public offering of stock

1989 - Teradata and NCR partner on next generation of DBC

1991 - NCR Corporation is acquired by AT&T- Teradata revenues at $280 million

1992 - Teradata is merged into NCR

1996 - AT&T spins off NCR Corp. with Teradata product

1997 - Teradata database becomes industry leader in data warehousing

2000 - 100+ Terabyte system in production

2003 - Teradata V2R5 released 12/20/2002; major release including features such asPPI, roles and profiles, multi-value compression, and more.



How Large is a Trillion?Teradata was the first commercial database system to support a trillion bytes of data. It ishard to imagine the size of a trillion. Most people are comfortable with Kilobytes,Megabytes, and even Gigabytes. However, Terabytes are another order of magnitude. Toput it in perspective, the life span of the average person is 2.5 Gigaseconds (or saiddifferently 2,500,000,000 seconds). A tera-second is 31,688 years!

1 Kilobyte = 103 = 1000 bytes1 Megabyte = 106 = 1,000,000 bytes1 Gigabyte = 109 = 1,000,000,000 bytes1 Terabyte = 1012 = 1,000,000,000,000 bytes1 Petabyte = 1015 = 1,000,000,000,000,000 bytes1 Exabyte = 1018 = 1,000,000,000,000,000,000 bytes1 Zetabyte = 1021 = 1,000,000,000,000,000,000,000 bytes1 Yottabyte = 1024 = 1,000,000,000,000,000,000,000,000 bytes


How Large is a Trillion?

1 Kilobyte = 103 = 1000 bytes1 Megabyte = 106 = 1,000,000 bytes1 Gigabyte = 109 = 1,000,000,000 bytes1 Terabyte = 1012 = 1,000,000,000,000 bytes1 Petabyte = 1015 = 1,000,000,000,000,000 bytes

1 million seconds = 11.57 days1 billion seconds = 31.6 years1 trillion seconds = 31,688 years

1 million inches = 15.7 miles1 trillion inches = 15,700,000 miles (30 roundtrips to the moon)

1 million square inches = .16 acres = .0002 sq. miles1 trillion square inches = 249 square miles (larger than Singapore)

$1 million = < $ .01 for every person in U.S.$1 billion = $ 3.64 for every person is U.S.$1 trillion = $ 3,636 for every person in U.S.



The Teradata CharterTeradata’s founders surveyed potential business partners/customers to determine their needsand requirements:

Relational – The relational model has become the accepted standard for database design. Itprovides the flexibility needed to leverage business information.

Large capacity database machine – data storage requirements are larger as a result of theneed for processing large amounts of detail data for decision support.

• Billions of Rows• Terabytes of data

Performance – as table size increased, early relational systems suffered severe performancelimitations.

Single data store for multiple clients – instead of replicating a database for different hosts,store it once, and use it from all clients.

Network connectivity – to connect easily to network attached host systems.Standard access language (SQL) – SQL has been adapted as the industry standard forrelational databases.

Manageable growth – we wanted a system that is linearly expandable to allow for growthwithout performance drop-off.

Fault tolerance – the ability to automatically detect and recover from one or more hardwarefailures.

Data integrity – Transactions are either completed or, if a fault occurs, rolled back, thusguaranteeing the integrity of the data.

The Teradata charter presented a challenge. Throughout this course, you will see howTeradata has met this challenge.


The Teradata Charter

• Relational database

• Enormous capacity

– Billions of rows

– Terabytes of data

• High performance parallel processing

• Single database server for multiple clients

• Network and mainframe connectivity

• Utilize an industry standard access language - Structured Query Language

– SQL (Structured Query Language)

• Manageable growth via modularity

• Fault tolerance at all levels of hardware and software

• Data integrity and reliability



Teradata’s Competitive AdvantagesThe facing page lists a number of the key competitive advantages that Teradata provides.This course will look at these features in detail and explain why these are competitiveadvantages.

Teradata provides a central, enterprise-wide database that contains information extractedfrom operational data stores. It provides for a single version of the truth. Characteristicsinclude:

• Based on enterprise-wide model – this type of model provides the ability tolook/work across functional processes.

• Customers can begin small (right size), but may grow large rapidly• Populated by extraction/loading of data from operational systems• Allows end-users to submit “what if” queries

Examples of applications that Teradata enables include:

• Customer Relationship Management (CRM)• Campaign Management• Yield Management• Supply Chain Management

Some of the reasons that Teradata is the leader in data warehousing include:

• Scalable – supports a small (10 GB) to a massive (100+TB) database.

• Provides a query optimizer with approximately 20 years of experience in large-tablequery planning.

• Does not require complex indexing schemes, complex data partitioning or time-consuming reorganizations (re-orgs).

• Supports ad hoc querying against the detail data in the warehouse, not just summarydata in the data mart.

• Designed and built with parallelism from day one (not a parallel retrofit).


Teradata's Competitive Advantages

• Unlimited, Proven Scalability - amount of data and number of users; allowsfor an enterprise wide model of the data.

• Unlimited Parallelism - Parallel access, sorts, and aggregations.

• Mature Optimizer - Handles complex queries, up to 64 joins per query, ad-hocprocessing.

• Models the Business - 3NF, robust view processing, & provides star schemacapabilities.

• Provides a “single version of the truth”.

• Low TCO (Total Cost of Ownership) - ease of setup, maintenance, &administration; no re-orgs, lowest disk to data ratio, and robust expansionutility (reconfig).

• High Availability - no single point of failure.

• Parallel Load utilities - robust, parallel, and scalable load utilities such asFastLoad, MultiLoad, and TPump.



Review QuestionsCheck your understanding of the concepts discussed in this module by completing thereview questions as directed by your instructor.


Review Questions

1. Name the two primary operating systems that the Teradata RDBMS executes on.______________________________________________

2. Which of the following represents a trillion bytes or a TB of data? ____

a. 106

b. 109

c. 1012

d. 1015

3. Which feature allows Teradata to process enormous volumes of data quickly? ____

a. High availability software and hardware componentsb. Parallelismc. Proven Scalabilityd. High performance servers from Intel

4. The Teradata RBDMS is primary a ____ .

a. Serverb. Client



Notes

Relational Database Concepts Page 2-1

Module 2


Relational Database Concepts


• Define the terms associated with databases

• Define the terms associated with relational databases

• List the advantage of a relational database


Page 2-2 Relational Database Concepts

Notes


Table of Contents

WHAT IS A DATABASE? ..................................................................................................................................... 2-4RELATIONAL DATABASES................................................................................................................................ 2-6PRIMARY KEY ...................................................................................................................................................... 2-8

PRIMARY KEY RULES ............................................................................................................................................ 2-8SELECTING A PRIMARY KEY .................................................................................................................................. 2-8

FOREIGN KEY ..................................................................................................................................................... 2-10FOREIGN KEY RULES ........................................................................................................................................... 2-10EXERCISE: ANSWERING QUESTIONS WITH A RELATIONAL DATABASE ................................................................ 2-12

ADVANTAGES OF A RELATIONAL DATABASE APPROACH ................................................................. 2-14REVIEW QUESTIONS......................................................................................................................................... 2-16



What is a Database?A database is a collection of permanently stored data that is used by an application orenterprise.

A database contains logically related data. Basically, that means that the database wascreated with a purpose in mind. A database supports shared access by many users. One ofthe characteristics of a database is that many people use it, often for many differentpurposes. A database also is protected to control access and managed to retain its valueand integrity.

One example of a database is payroll data that includes the name of the employee, theiremployee number, and their salary history. This database is logically related – it’s all aboutpayroll. It certainly must have shared access, since it will be used by the payroll departmentto generate checks, and also by management to make decisions. This database must also beprotected – much of the information is confidential – and managed to ensure the accuracy ofthe records.

Later in this course we will provide another definition of “database” that is specific toTeradata.


What is a Database?

Database – A collection of permanently stored data that is:

• Logically related (data relates to other data)

• Shared (many users may access data)

• Protected (access to data is controlled)

• Managed (data has integrity and value)

TeradataDATABASE

Win 2000

IBMMainframe

UNIX

Win 98



Relational DatabasesThe key to understanding relational databases is the concept of the table made up of rowsand columns.

A column always contains like data. In the example on the following page, the columnnamed LAST NAME contains last name, and never anything else. The position of thecolumn in the table is arbitrary.

A row is one instance of all the columns of a table. In our example, all of the informationabout a single employee is in one row. The sequence of the rows in a table is arbitrary.

Specifically, in a Relational Database, tables are defined as a named collection of one ormore named columns by zero or more rows of related information.

Notice that each row of the table is about a person. There are no rows with data on twopeople, nor are there rows with information on anything other than people. This may seemobvious, but the concept underlying it is very important.

Each row represents an occurrence of an entity defined by the table. An entity is defined asa person, place or thing about which the table contains information. In this case the entity isthe employee.


Relational Databases

A Relational Database consists of a set of logically related tables.

A table is a two dimensional representation of data consisting of rows andcolumns.

The employee table has nine columns of data.It has six rows of data - one per employee.There is no prescribed order for the rows of the table.There is only one row “format” for the entire table.Missing data values are represented by “nulls”.

1006 1019 301 312101 Stein John 861015 631015 39450001008 1019 301 312102 Kanieski Carol 870201 680517 39250001005 0801 403 431100 Ryan Loretta 861015 650910 41200001004 1003 401 412101 Johnson Darlene 861015 560423 46300001007 1005 403 432101 Villegas Arnando 870102 470131 59700001003 0801 401 411100 Trader James 860731 570619 4785000

EMPLOYEEColumn

MANAGEREMPLOYEE EMPLOYEE DEPT JOB LAST FIRST HIRE BIRTH SALARYNUMBER NUMBER NUMBER CODE NAME NAME DATE DATE AMOUNT

Row



Primary KeyTables, made up of rows and columns, represent entities or relationships. Entities are thepeople, places, things, or events that the Entity Tables Model. Each table holds only onekind of row, and each row is uniquely identified within a table by a Primary Key (PK).

Primary Key RulesA Primary Key is required. All tables must have one, because each row within a table mustbe able to be uniquely identified. A Primary Key uniquely identifies each row in a table.

No duplicate values are allowed. The Primary Key for the EMPLOYEE table is theEmployee number. No two employees can have the same number.

Because it is used to identify, the Primary Key cannot be null. There must be something inthat field to uniquely identify each occurrence.

Primary Key values cannot be changed. Historical information as well as relationshipswith other entities may be lost if a PK value is changed or re-used.

A Primary Key can be more than one column.

Only one Primary Key is allowed per table.

Selecting a Primary KeyEach column of a row is called an attribute of that row. A database designer can select anyattribute to be a Primary Key, but as a result of the rules listed above, many attributes willnot qualify as a Primary Key candidate. For example, we could have selected Last Name asthe PK of the EMPLOYEES table, but as soon as the company hires two people with thesame last name the PK is no longer unique. Even if we made the PK Last Name and FirstName (possible since PKs can be made up of more than one column) we could still havetwo employees with the same name. Moreover, some employees may choose to change theirlast name.

Many data modelers recommend using system-assigned sequential integers for PrimaryKeys. This assures uniqueness and gives the users control over the Primary Key values.


Primary Key

Primary Key values uniquely identify each row in a table.


EMPLOYEEMANAGER

EMPLOYEE EMPLOYEE DEPT JOB LAST FIRST HIRE BIRTH SALARYNUMBER NUMBER NUMBER CODE NAME NAME DATE DATE AMOUNT

PK

In a relational model,

• A Primary Key is required for every table.

• Only one Primary Key is allowed in a table.

• It may consist of one or more columns.

• Primary Keys cannot have duplicate values.

• Primary Keys cannot be NULL.

• Primary Keys are considered “non-changing” values.



Foreign KeyRelational Databases permit associations by data value across more than one table. ForeignKeys (FKs) model the relationships between entities.

On the facing page you will see that the employee table has 3 FK columns, one of whichmodels the relationship between employees and their departments. A second one models therelationship between employees and their job codes.

A third FK column is used to model the relationship between employees and each other.This is called a “recursive” relationship.

Foreign Key RulesRules of Foreign Keys include:

• Duplicate values are allowed in a FK column.

• Missing values are allowed in a FK column.

• Values may be changed in a FK column.

• Each FK value must exist as a Primary Key.

Note that Department_Number is the Primary Key for the DEPARTMENT table.


Foreign Key

Foreign Key (FK) values identify table relationships.


EMPLOYEE (partial listing)MANAGER


PK FK FK FK

MANAGERDEPT DEPARTMENT BUDGET EMPLOYEENUMBER NAME AMOUNT NUMBER

501 marketing sales 80050000 1017301 research and development 46560000 1019302 product planning 22600000 1016403 education 93200000 1005402 software support 30800000 1011401 customer support 98230000 1003201 technical operations 29380000 1025

PK FK

DEPARTMENT

• FK’s are optional - not all tableshave them

• More than one FK is allowed pertable

• FK’s can be made up of more thanone column

• Duplicate values are allowed• Missing (NULL) values are allowed• Changes are allowed• Each FK value must exist somewhere

as a PK value



Exercise: Answering Questions with a Relational DatabaseA relational database is a collection of relational tables stored in a single installation of arelational database management system (RDBMS). The words “management system”indicate that not only is this a relational database but also there is underlying software toprovide additional functions that the industry expects. This includes transaction integrity,security, journaling, and other features that are expected of databases in general. TheTeradata Database is a Relational Database Management System.

Relational databases do not use access paths to locate data, rather data connections are madeby data values. In other words, data connections are made by matching values in onecolumn with the values in a corresponding column in another table. This connection isreferred to as a JOIN in relational terminology.

The diagram on the facing page show how the values in one table may be matched to valuesin another.

Both tables have a column named “Department Number”. That connection allows thedatabase to answer questions like, “What is the name of the department in which anemployee works?”

One reason relational databases are so powerful is that, unlike other databases, they arebased on a mathematical model developed by Dr. Edgar Codd and implement a querylanguage solidly founded in set theory.

To sum up, a relational database is a collection of tables. The data contained in the tablescan be associated using data values, specifically, columns with matching data values.


Exercise: Answering Questions with aRelational Database




PK FK FK FK

MANAGERDEPT DEPARTMENT BUDGET EMPLOYEENUMBER NAME AMOUNT NUMBER


PK FK

DEPARTMENT

Questions:1. Name the department in which

James Trader works.2. Who manages the Education

Department?3. Identify by name an employee

who works for James Trader.4. James Trader manages which

department?



Advantages of a Relational Database ApproachRelational Databases are more flexible than other databases. There are two reasons for thisimportant feature. First, a relational database does not require the user to know the accesspath. The RDBMS keeps track of where everything is. Second, relational databases useatomic data. Breaking data down into its smallest atomic parts allows maximum flexibilityin selecting and using data.

Relational data allows businesses to respond to changing conditions more quickly. In atraditional database, adding a field meant that all the programs that used the database had tobe rewritten to be aware of the new data structure. That is not true in a relational DBS.

Relational systems are data-driven, not application-driven. Relational databases aredesigned to represent the business and the business practices, not the application or thecomputer system.

Relational systems are business-oriented. The two tables we've looked at, EMPLOYEE,and DEPARTMENT, are organized to reflect the way the business really works.

Relational systems are designed to be simple and easy to use. This feature is useful notonly to the people who ask the questions, but also to the people who have to figure out howto get the information out of the database.

The last point we discussed is directly related to the next one. Relational systems make thedata do more work. Programs and transactions become much simpler. And that makes iteasier to build applications.

Relational Theory is easy to understand. Understanding how a RDBMS does what it doesis not necessary, any more than it's necessary to know the actual gear ratios of yourautomatic transmission. All the user needs to know is what it can do.

In the 1990's, the trend is away from organizations funneling all data requests through a fewpeople who know how the system works. As systems get easier to use, more people haveaccess to them. This is called the “democratization of data”. It is important to note thatrelational databases support this trend toward end-user computing.


Advantages of a Relational Database Approach

The advantages of a Relational Database compared to other databasemethodologies are many.

Relational database methodology:

• Is easy to use

• Is easy to understand

• Models the business, not the processes

• Is data-driven versus application driven

• Makes applications easier to build

• Supports trend toward end-user computing

• Is the industry standard for most large enterprises

• Allows businesses to respond to changing conditions more flexibly thanother types


Review Questions

Match each term with its definition below:

___ 1. Database

___ 2. Table

___ 3. Relational database

___ 4. Primary Key

___ 5. Null

___ 6. Foreign Key

a - A set of columns which uniquely identify a rowb - A set of logically related tablesc - One or more columns that are a PK somewhere in the databased - The absence of a valuee - A two-dimensional array of rows and columnsf - A collection of permanently stored data



Notes

Teradata Basics Page 3-1

Module 3


Teradata Basics


• List and describe the major components of the Teradataarchitecture.

• Describe how the components interact to manageincoming and outgoing data

• List the 5 types of Teradata database objects


Page 3-2 Teradata Basics

Notes


Table of Contents

TERADATA STORAGE ARCHITECTURE ....................................................................................................... 3-4TERADATA RETRIEVAL ARCHITECTURE................................................................................................... 3-6MULTIPLE TABLES ON MULTIPLE AMPS .................................................................................................... 3-8

Here's how it works:......................................................................................................................................... 3-8LINEAR GROWTH AND EXPANDABILITY .................................................................................................. 3-10TERADATA OBJECTS........................................................................................................................................ 3-12

Tables ............................................................................................................................................................. 3-12Views .............................................................................................................................................................. 3-12Macros............................................................................................................................................................ 3-12Triggers .......................................................................................................................................................... 3-12Stored Procedures .......................................................................................................................................... 3-12

THE DATA DICTIONARY DIRECTORY (DD/D) ........................................................................................... 3-14STRUCTURE QUERY LANGUAGE (SQL)...................................................................................................... 3-16

DATA DEFINITION LANGUAGE (DDL) ................................................................................................................. 3-16DATA MANIPULATION LANGUAGE (DML) .......................................................................................................... 3-16DATA CONTROL LANGUAGE (DCL) .................................................................................................................... 3-16USER ASSISTANCE ............................................................................................................................................... 3-16

CREATE TABLE – EXAMPLE OF DDL........................................................................................................... 3-18VIEWS.................................................................................................................................................................... 3-20

SINGLE-TABLE VIEW............................................................................................................................................ 3-20MULTI-TABLE VIEWS....................................................................................................................................... 3-22SELECT – EXAMPLE OF DML......................................................................................................................... 3-24

THE SELECT STATEMENT .................................................................................................................................. 3-24THE JOIN OPERATION ..................................................................................................................................... 3-26MACROS – TERADATA SQL EXTENSION .................................................................................................... 3-28MACROS................................................................................................................................................................ 3-28

FEATURES OF MACROS ........................................................................................................................................ 3-28BENEFITS OF MACROS ......................................................................................................................................... 3-28

HELP COMMANDS – TERADATA SQL EXTENSION.................................................................................. 3-30EXAMPLE OF HELP DATABASE ....................................................................................................................... 3-32

SHOW COMMAND – TERADATA SQL EXTENSION .................................................................................. 3-34EXPLAIN FACILITY – TERADATA SQL EXTENSION ............................................................................... 3-36TERADATA FEATURES REVIEW ................................................................................................................... 3-38REVIEW QUESTIONS......................................................................................................................................... 3-40



Teradata Storage ArchitectureUp until now we have discussed relational databases in terms of how the user perceivesthem – as a collection of tables that relate to one another. Now it's time to describe thephysical components of the system.

On the facing page you will see a simplified view of how the physical components of aTeradata database work to insert a row of data.

The Parsing Engine interprets the SQL command and converts the data record from thehost into an AMP message.

The Message Passing Layer distributes the row to the appropriate Access ModuleProcessor (AMP).

The AMP formats the row and writes it to its associated disks.

The disk holds the row for subsequent access.

The Host or Client system supplies the records. These records are the raw data from whichthe database will be constructed.

The Parsing Engine is a component that interprets SQL requests, receives input records andpasses data. To do that it sends the messages through the Message Passing Layer to theAMPs.

The Message Passing Layer is implemented variously as hardware or software, dependingon the platform used. It determines which unit or units should receive a message.

Think of the AMP (Access Module Processor) as a computer designed for and dedicated tomanaging a portion of the entire database. It performs all the database managementfunctions – such as sorting, aggregating, and formatting the data. It receives data from thePE, formats the rows, and distributes the rows to the disk storage units it controls. It alsoretrieves the rows requested by the parsing engine.

Disks are simply disk drives associated with an AMP. They store the data rows.On current systems, they are usually implemented using a disk array (to be discussed later).


Teradata Storage Architecture

Notes:

The Parsing Engine dispatchesrequest to insert a row.

The Message Passing Layerinsures that a row gets to theappropriate AMP (Access ModuleProcessor).

The AMP stores the row on itsassociated (logical) disk.

An AMP manages a logical orvirtual disk which is mapped tomultiple physical disks in a diskarray.

Teradata

AMP 4AMP 3AMP 1 AMP 2

ParsingEngine(s)

Message Passing Layer

182 54

41

1290

75

80

32 667

25

Records From Client (in random sequence)2 32 67 12 90 6 54 75 18 25 80 41



Teradata Retrieval ArchitectureRetrieving data from the Teradata RDBMS simply reverses the process of the storagemodel. A request is made for data and is passed on to a Parsing Engine (PE). The PEoptimizes the request for efficient processing and creates tasks for the AMPs to perform,which will result in the request being satisfied. These tasks are then dispatched to the AMPsvia the Message Passing Layer. Often times all AMPs must participate in creating theanswer set, such as in returning all rows of a table. Other times, only one or a few AMPsneed participate, depending on the nature of the request. The PE will insure that only theAMPs that are needed will be assigned tasks on behalf of this request.

Once the AMPs have been given their assignments, they will retrieve the desired rows fromtheir respective disks. If sorting, aggregating or formatting of any kind is needed, the AMPswill also take care of that. The rows are then returned to the requesting PE via the MessagePassing Layer. The PE takes the returned answer set and returns it to the requesting clientapplication.


Teradata Retrieval Architecture

Notes:

The Parsing Engine dispatches arequest to retrieve one or morerows.

The Message Passing Layerinsures that the appropriateAMP(s) are activated.

The AMP(s) locate and retrievedesired row(s) in parallel access.

Message Passing Layer returns toretrieved rows to PE.

The PE returns row(s) torequesting client application.

Teradata

AMP 4AMP 3AMP 1 AMP 2

ParsingEngine(s)


182 54

41

1290

75

80

32 667

25

Rows retrieved from table2 32 67 12 90 6 54 75 18 25 80 41



Multiple Tables on Multiple AMPsLogically, you might think that the RDBMS would assign each table to a particular AMP,and that the AMP would put that table on a single disk. However, as you see on the diagramon the facing page, that's not what happens. The system takes the rows that composes atable and divides those rows up among all available AMPs.

Here's how it works:Tables are distributed across all AMPs. This distribution of rows should be even across allAMPs. This way, a request to get the rows of a given table will result in the workload beingevenly distributed across the AMPs.

• Each table has some rows distributed to each AMP.

• Each AMP controls one logical storage unit which may consist of several physicaldisks

• Each AMP places, maintains, and manages the rows on its own disks.• Large configurations may have hundreds of AMPs.

• Full table scans, operations that require looking at all the rows of a table, access allAMPs in parallel. That parallelism is what makes possible the accessing ofenormous amounts of data.

Consider the following three tables: EMPLOYEE, DEPARTMENT, and JOB.

The TERADATA RDBMS takes the rows from each of the tables and divides them upamong all the AMPs. The AMPs divide the rows up among their disks. Notice that eachAMP gets part of each table. Dividing up the tables this way means that all the AMPs andtheir associated disks will be activated in a full table scan, thus speeding up requests againstthese tables.

In our example, if you assume four AMPs, each AMP would get approximately 25% ofeach table. If, however, AMP #1 were to get 90% of the rows from the EMPLOYEE tablethat would be called "lumpy" data distribution. Lumpy data distribution would slow thesystem down because any request that required scanning all the rows of EMPLOYEE wouldhave three AMPs sitting idle while AMP #1 finished its work. It is better to divide all thetables up evenly among all the available AMPs. You will see how this distribution iscontrolled in a later chapter.


Multiple Tables on Multiple AMPs

EMPLOYEE RowsDEPARTMENT RowsJOB Rows

EMPLOYEE Table DEPARTMENT Table JOB Table

Parsing Engine

AMP #1 AMP #2 AMP #3 AMP #4


Notes:

Some rows from each table maybe found on each AMP.

Each AMP may have rows fromall tables.

Ideally, each AMP will holdroughly the same amount ofdata.






Linear Growth and ExpandabilityThe Teradata DBS is the first commercial database system to offer true parallelism and theperformance increase that goes with it.

Think back to the example of how rows are divided up among AMPs that we just discussed.Assume that our three tables, EMPLOYEE, DEPARTMENT, and JOB total 100,000 rows,with a certain number of users, say 50. What happens if you double the number of AMPs and the number of users stays the same?It doubles. Each AMP can only work on half as many rows as they used to.

Now think of that system in a situation where the number of users is doubled, as well as thenumber of AMPs. We now have 100 users, but we also have twice as many AMPs. Whathappens to performance? It stays the same. There is no drop-off in the speed with whichrequests are executed.

That's because the system is modular and the workload is easily partitioned intoindependent pieces. In the last example, each AMP is still doing the same amount of work.

This feature -- that the amount of time (or money) required to do a task is directlyproportional to the size of the system -- is unique to the Teradata RDBMS. Traditionaldatabases show a sharp drop in performance when the system approaches a critical size.

Look at the diagram on the facing page. As the number of Parsing Engines increases, thenumber of SQL requests that can be supported increases.

As you add AMPs, data is spread out more even as you add processing power to handle thedata.

As you add disks, you add space for each AMP to store and process more information. AllAMPs must have the same amount of disk storage space.


Linear Growth and Expandability

AMPAMP

ParsingEngine

AMP

SESSIONS

PARALLEL PROCESSING

DATADisk

DiskDisk

ParsingEngine

ParsingEngine Notes:

Teradata is a linearlyexpandable DBMS.

Components may be added asrequirements grow.

Performance impact of addingcomponents is shown below.

USERS AMPs DATA PerformanceSame Same Same SameDouble Double Same SameSame Double Double SameSame Double Same Double



Teradata ObjectsA “database” in Teradata database systems is a collection of objects known as tables,views, macros, triggers, or stored procedures. All database objects are created andaccessed using the standard Structured Query Language or SQL.

All database object definitions are stored in a system database called the DataDictionary/Directory (DD/D).

Databases provide a logical grouping for information. They are also the foundation forspace allocation and access control.

TablesA table is the logical structure of data in an RDBMS. It is a two-dimensional structuremade up of columns and rows. A user defines a table by giving it a table name that refersto the type of data that will be stored in the table.

A column represents attributes of the table. Column names are given to each column of thetable. All the information in a column is the same type, for example, date of birth.

Each occurrence of an entity is stored in the table as a row. Entities are the people, things,or events that the table is about. Thus a row would represent a particular person, thing, orevent.

ViewsA view is a pre-defined subset of one of more tables or other views. It does not exist as areal table, but serves as a reference to existing tables or views. One way to think of a viewis as a virtual table. Views have definitions in the data dictionary, but do not contain anyphysical rows. The database administrator can use views to control access to the underlyingtables. Views can be used to hide columns from users, to insulate applications fromdatabase changes, and to simplify or standardize access techniques.

MacrosA macro is a predefined, stored set of one or more SQL commands and optionally, reportformatting commands. Macros are used to simplify the execution of frequently used SQLcommands.

TriggersA trigger is a set of SQL statements usually associated with a column or a table and whenthat column changes, the trigger is fired – effectively executing the SQL statements.

Stored ProceduresA stored procedure is a program that is stored within Teradata and executes within theTeradata RDBMS. A stored procedure uses permanent disk space.


Teradata ObjectsThere are eight types of objects which may be found in a Teradata database/user.

Tables – rows and columns of dataViews – predefined subsets of existing tablesMacros – predefined, stored SQL statementsTriggers – SQL statements associated with a tableStored Procedures – program stored within TeradataJoin and Hash Indexes – separate index structures stored as objects within a databasePermanent Journals – table used to store before and/or after images for recovery

DEFINITIONS OFALL DATABASE

OBJECTS

DD/D

These objects are created,maintained and deleted usingStructured Query Language (SQL).

Object definitions are stored in theData Dictionary / Directory (DD/D).

DATABASE or USER

TABLE 2 TABLE 3TABLE 1

VIEW 2 VIEW 3VIEW 1

MACRO 2 MACRO 3MACRO 1

TRIGGER 2 TRIGGER 3TRIGGER 1

Stored Procedure 1 Stored Procedure 2 Stored Procedure 2

Join/Hash Index 1 Join/Hash Index 2 Join/Hash Index 3

Permanent Journal These aren't directly accessed by users.



The Data Dictionary Directory (DD/D)The Data Dictionary/Directory is an integrated set of system tables which store databaseobject definitions and accumulate information about users, databases, resource usage,data demographics, and security rules. It records specifications about tables, views, andmacros. It also contains information about ownership, space allocation, accounting, andaccess rights (privileges) for these objects.

Data Dictionary/Directory information is updated automatically during the processing ofTeradata SQL data definition (DDL) statements. It is used by the Parser to obtaininformation needed to process all Teradata SQL statements.

Users may access the DD/D through Teradata-supplied views, if permitted by the systemadministrator.


The Data Dictionary Directory (DD/D)

The DD/D ...

– is an integrated set of system tables

– contains definitions of and information about all objects in the system

– is entirely maintained by the RDBMS

– is “data about the data” or “metadata”

– is distributed across all AMPs like all tables

– may be queried by administrators or support staff

– is accessed via Teradata supplied views

Examples of DD/D views:

DBC.Tables - information about all tables

DBC.Users - information about all users

DBC.AllRights - information about access rights

DBC.AllSpace - information about space utilization



Structure Query Language (SQL)Structured Query Language (SQL) is the language of relational databases. It is sometimesreferred to as a "Fourth Generation Language (4GL)" to differentiate it from "ThirdGeneration Languages" such as FORTRAN and COBOL, though it is quite different fromother 4GL’s. It acts as an intermediary between the user and the database.

SQL is different in some very important ways from other computer languages. Itsstatements resemble English-like structures. It provides powerful, set-oriented databasemanipulation including structural modification, data retrieval, modification, and securityfunctions.

SQL is a non-procedural language. Because of its set orientation it does not require IF,GOTO, DO, FOR NEXT or PERFORM statements.

We'll describe three important subsets of SQL – the Data Definition Language, the DataManipulation Language, and the Data Control Language.

Data Definition Language (DDL) The DDL allows a user to define the database objects and the relationships that existamong them. Examples of DDL uses are creating or modifying tables and views.

Data Manipulation Language (DML)The DML consists of the statements that manipulate, change or retrieve the data rowsof the database. If the DDL defines the database, the DML lets the user change theinformation contained in the database. The DML is the most commonly used subset ofSQL. It is used to select, update, delete, and insert rows.

Data Control Language (DCL)The Data Control Language is used to restrict or permit a user's access in various ways. Itcan selectively limit a user's ability to retrieve, add, or modify data. It is used to grant andrevoke access privileges on tables and views. An example is granting update privileges on atable, or read privileges on a view to specified users.

User AssistanceThese commands allow you to list the objects in a database, or the characteristics of a table,see how a query will execute, or show you the details of your system. They vary widelyfrom vendor to vendor.


Structured Query Language (SQL)

SQL is a query language for Relational Database Systems.– A fourth-generation language– A set-oriented language– A non-procedural language

(e.g, doesn’t have IF, GO TO, DO, FOR NEXT, or PERFORM statements)

SQL consists of:Data Definition Language (DDL)

– Defines database structures (tables, users, views, macros, triggers, etc.)

CREATE DROP ALTER

Data Manipulation Language (DML)– Manipulates rows and data values

SELECT INSERT UPDATE DELETE

Data Control Language (DCL)– Grants and revokes access rights

GRANT REVOKE

Teradata SQL also includes Teradata Extensions to SQL

HELP SHOW EXPLAIN CREATE MACRO



CREATE TABLE – Example of DDLTo create and store the table structure definition in the DD/D, you can execute the CREATETABLE DDL statement as shown on the facing page.

An example of the output from a SHOW TABLE command follows:

SHOW TABLE Employee;

CREATE SET TABLE Per_DB.Employee, FALLBACK , NO BEFORE JOURNAL, NO AFTER JOURNAL, DATABLOCKSIZE = 16384 BYTES, FREESPACE = 30 PERCENT ( employee_number INTEGER NOT NULL,

manager_emp_number INTEGER NOT NULL, dept_number SMALLINT, job_code INTEGER COMPRESS , last_name CHAR(20) NOT CASESPECIFIC NOT NULL, first_name VARCHAR(20) NOT CASESPECIFIC, hire_date DATE FORMAT 'YYYY-MM-DD' birth_date DATE FORMAT 'YYYY-MM-DD', salary_amount DECIMAL(10,2) )

UNIQUE PRIMARY INDEX ( employee_number )INDEX ( dept_number );

You can create secondary indexes after a table has been created by executing the CREATEINDEX command. An example of creating an index for the job_code column is shown onthe facing page.

Examples of the DROP INDEX and DROP TABLE commands are also shown on the facingpage.


CREATE TABLE – Example of DDL

CREATE TABLE Employee,FALLBACK(employee_number INTEGER NOT NULL,manager_emp_number INTEGER,dept_number SMALLINT,job_code INTEGER COMPRESS,last_name CHAR(20) NOT NULL,first_name VARCHAR (20),hire_date DATE FORMAT 'YYYY-MM-DD',birth_date DATE FORMAT 'YYYY-MM-DD',salary_amount DECIMAL (10,2)) UNIQUE PRIMARY INDEX (employee_number),INDEX (dept_number);

Other DDL Examples

CREATE INDEX (job_code) ON Employee ;

DROP INDEX (job_code) ON Employee ;

DROP TABLE Employee ;



ViewsA view is a pre-defined subset of one or more tables. Views are used to control access to theunderlying tables and simplify access to data. Authorized users may use views to read dataspecified in the view and/or to update data specified in the view.

Views are used to simplify query requests, to limit access to data, and to allow differentusers to look at the same data from different perspectives.

A view is a window that accesses selected portions of a database. Views can show parts ofone table (single-table view), more than one table (multi-table view), or a combination oftables and other views. To the user, views look just like tables.

Views are an alternate way of organizing and presenting information. A view, like atable, has rows and columns. However, the rows and columns of a view are not storeddirectly but are derived from the rows and columns of tables whenever the view isreferenced. A view looks like a table, but has no data of its own, and therefore takes up nostorage space except for its definition. One way to think of a view is as if it was a windowthrough which you can look at selected portions of a table or tables.

Single-table ViewA single-table view takes specified columns and/or rows from a table and makes themavailable in a fashion that looks like a table. An example might be an employee table fromwhich you select only certain columns for employees in a particular department number, forexample, department 403, and present them in a view.

Example of a CREATE VIEW statement:

CREATE VIEW Emp_403 ASSELECT employee_number

,department_number,last_name,first_name,hire_date

FROM EmployeeWHERE department_number = 403;

It is also possible to execute SHOW VIEW viewname.


Views

Views are pre-defined subsets of existing tables consisting of specifiedcolumns and/or rows from the table(s).

A single table view:– is a window into an underlying table– allows users to read and update a subset of the underlying table– has no data of its own

MANAGEREMPLOYEE EMP DEPT JOB LAST FIRST HIRE BIRTH SALARYNUMBER NUMBER NUMBER CODE NAME NAME DATE DATE AMOUNT


EMPLOYEE (Table)

PK FK FK FK

EMP NO DEPT NO LAST NAME FIRST NAME HIRE DATE

1005 403 Villegas Arnando 870102 801 403 Ryan Loretta 861015

Emp_403 (View)



Multi-Table ViewsA multi-table view combines data from more than one table into one pre-defined view.These views are also called “join views” because more than one table is involved.

An example might be a view that shows employees and the name of their department,information that comes from two different tables.

Note: Multi-table Views are read only. The user cannot alter the data via the view.

One might wish to create a view containing the last name and department name for allemployees.

Example of SQL to create a join view:

CREATE VIEW EmpDept ASSELECT last_name

, department_nameFROM Employee E INNER JOIN Department DON E.department_number = D.department_number ;

An example of reading via this view is:

SELECT last_name,department_name

FROM EmpDept;

This example utilizes an alias name of E for the Employee table and D for the Departmenttable.


Multi-Table Views

A multi-table view allows users to access data from multiple tables as if it were in a singletable. Multi-table views are also called join views. Join views are used for reading only,not updating.

EMPLOYEE (Table)



PK FK FK FK

MANAGERDEPT DEPARTMENT BUDGET EMPNUMBER NAME AMOUNT NUMBER


PK FK

DEPARTMENT (Table)

LAST DEPARTMENT NAME NAME

Stein research & developmentKanieski research & developmentRyan educationJohnson customer supportVillegas educationTrader customer support

EmpDept (View)

"Joined Together"



SELECT – Example of DML

The SELECT StatementSELECT is the most commonly used SQL statement. It is a data manipulation statementthat returns information from the database and organizes it for presentation to the user orto the application program.

Specifically, SELECT allows the user to return data from one or more tables.

The following SELECT statement is an example of selecting data from one table.

SELECT Last_name, First_nameFROM EmployeeWHERE Hire_date = '1986-10-15’;

On the facing page you will see how this statement answers the question: “Who was hiredon 10/15/86?”

This was an example of a SELECT on one table. What if you want information from morethan one table? That's our next topic.


SELECT – Example of DML

The SELECT statement is used to retrieve data from tables.

Who was hired on October 15, 1986?



EMPLOYEE EMP DEPT JOB LAST FIRST HIRE BIRTH SALARYNUMBER NUMBER NUMBER CODE NAME NAME DATE DATE AMOUNT

PK FK FK FK

SELECT Last_Name ,First_Name

FROM EmployeeWHERE Hire_Date = '1986-10-15';

Result

LASTNAMEStein RyanJohnson

FIRSTNAMEJohnLorettaDarlene



The JOIN OperationA Join operation joins rows of multiple tables and creates ‘virtual rows’. These are rowsthat contain data from more than one table but are not maintained anywhere in permanentstorage. The virtual rows are created dynamically as part of a join operation. Rows arematched up based on Primary and Foreign Key relationships.

Joins are an essential technique for accessing data in a relational database. Let’s look at oursample tables again and see how a Join would be used to answer the question: “Who worksin Research and Development?”

The SQL to accomplish this is a bit more complex and is covered in the Teradata SQLcourses.

Example of a JOIN statement:

SELECT E.first_name, E.last_nameFROM Employee E INNER JOIN Department DON E.department_number = D.department_numberAND D.department_name = ‘Research and Development’;

This example utilizes an alias name of E for the Employee table and D for the Departmenttable.


The JOIN Operation

A join operation is used when the SQL query requires information from multipletables.

Who works in Research and Development?EMPLOYEE



PK FK FK FK

MANAGERDEPT DEPARTMENT BUDGET EMPNUMBER NAME AMOUNT NUMBER


PK FK

DEPARTMENTResult

LASTNAMEStein Kanieski

FIRSTNAMEJohnCarol



Macros – Teradata SQL Extension

MacrosThe Macro facility allows you to define a sequence of Teradata SQL statements (andoptionally Teradata report formatting statements) so that they execute as a single transaction.Macros reduce the number of keystrokes needed to perform a complex task. This saves youtime, reduces the chance of errors, reduces the communication volume to Teradata, andallows efficiencies internal to Teradata.

Features of Macros• Macros are source code stored on the DBC.• They can be modified and executed at will.• They are re-optimized at execution time.• They can be executed by interactive or batch applications.• They are executed by one EXECUTE command.• They can accept user-provided parameter values.

Benefits of Macros• Macros simplify and control access to the system.• They enhance system security.• They provide an easy way of installing referential integrity.• They reduce the amount of source code transmitted from the client application. • They are stored in the Teradata DD/D and are available to all connected hosts.

To create a macro:

CREATE MACRO Customer_List AS(SELECT customer_name FROM Customer; );

To execute a macro:

EXEC Customer_List;

To replace a macro:

REPLACE MACRO Customer_List AS(SELECT customer_name, customer_number FROM Customer; );

To drop a macro:

DROP MACRO Customer_List;


Macros – Teradata SQL Extension

A MACRO is a predefined set of SQL statements which is logically stored in a database.

Macros may be created for frequently occurring queries of sets of operations.

Macros have many features and benefits:

• Simplify end-user access• Control which operations may be performed by users• May accept user-provided parameter values• Are stored on the RDBMS, thus available to all clients• Reduces query size, thus reduces LAN/channel traffic• Are optimized at execution time• May contain multiple SQL statements

To create a macro:CREATE MACRO Customer_List AS (SELECT customer_name FROM Customer;);

To execute a macro:

EXEC Customer_List;

To replace a macro:

REPLACE MACRO Customer_List AS (SELECT customer_name, customer_number FROM Customer;);



HELP Commands – Teradata SQL ExtensionHELP commands are available to display information on database objects:

• Databases and users• Tables• Views• Macros• Triggers• Stored Procedures


HELP Commands – Teradata SQL Extension

Databases and Users:

HELP DATABASE Customer_Service ;

HELP USER Dave_Jones ;

Tables, Views, and Macros:

HELP TABLE Employee ;

HELP VIEW Emp;

HELP MACRO Payroll_3;

HELP COLUMN Employee.*;

Employee.last_name;

Emp.* ;

Emp.last;

HELP INDEX Employee;

HELP STATISTICS Employee;

HELP CONSTRAINT Employee.over_21;



Example of HELP DATABASEThe facing page contains an example of a HELP DATABASE command. This commandlists the tables, views, macros, and triggers in the specified database.

The Kind column codes represent:

T – TableV – ViewM – MacroI – Join IndexP – Stored ProcedureN – Hash IndexJ – Permanent Journal


Example of HELP DATABASE

HELP DATABASE Customer_Service;*** Help information returned. 15 rows.*** Total elapsed time was 1 second.

Table/View/Macro name Kind CommentContact T ?Customer T ?Cust_Comp_Orders V ?Cust_Pend_Orders V ?Cust_Order_ix I ?Department T ?Employee T ?Employee_Phone T ?Job T ?Location T ?Location_Employee T ?Location_Phone T ?Orders T ?Set_Ansidate_on M ?Set_Integerdate_on M ?

Command:



SHOW Command – Teradata SQL ExtensionHELP commands display information about database objects (users/databases, tables, views,macros, triggers, and stored procedures) and session characteristics.

SHOW commands display the data definition (DDL) associated with database objects(tables, views, macros, triggers, or stored procedures).

BTEQ contains a SHOW command, in addition to and separate from the SQL SHOWcommand. The BTEQ SHOW provides information on the formatting and display settingsfor the current BTEQ session, if applicable.


SHOW Command – Teradata SQL Extension

SHOW commands display how an object was created.

Command Returns statementSHOW TABLE tablename; CREATE TABLE statementSHOW VIEW viewname; CREATE VIEW ...SHOW MACRO macroname; CREATE MACRO ...SHOW TRIGGER triggername; CREATE TRIGGER …SHOW PROCEDURE procedurename; CREATE PROCEDURE …

SHOW TABLE Employee;CREATE SET TABLE CUSTOMER_SERVICE.Employee ,FALLBACK ,NO BEFORE JOURNAL ,NO AFTER JOURNAL ( employee_number INTEGER, manager_employee_number INTEGER, department_number INTEGER, job_code INTEGER, : salary_amount DECIMAL(10,2) NOT NULL)UNIQUE PRIMARY INDEX ( employee_number );



EXPLAIN Facility – Teradata SQL ExtensionThe EXPLAIN facility allows you to preview how Teradata will execute a query you haverequested. It returns a summary of the steps the Teradata RDBMS would perform toexecute the request. EXPLAIN also discloses the strategy and access method to be used,how many rows will be involved, and its “cost” in minutes and seconds. You can useEXPLAIN to evaluate a query performance and to develop an alternative processingstrategy that may be more efficient. EXPLAIN works on any SQL request. The request isfully parsed and optimized, but it is not run. Instead, the complete plan is returned to theuser in readable English statements.

EXPLAIN also provides information about locking, sorting, row selection criteria, joinstrategy and conditions, access method, and parallel step processing.

There are a lot of reasons for using EXPLAIN. The main ones we’ve already pointed out –it lets you know how the system will do the job, what kind of results you will get back, andthe relative cost of the query. EXPLAIN is also useful for performance tuning, debugging,pre-validation of requests, and for technical training.

The following is an example of an EXPLAIN on a very simple query.

EXPLAIN SELECT last_name, department_number FROM Employee;

Explanation (full)---------------------------------------------------------------------------1) First, we lock a distinct CUSTOMER_SERVICE."pseudo table" for read on a

RowHash to prevent global deadlock for CUSTOMER_SERVICE.Employee.2) Next, we lock CUSTOMER_SERVICE.Employee for read.3) We do an all-AMPs RETRIEVE step from CUSTOMER_SERVICE.Employee by

way of an all-rows scan with no residual conditions into Spool 1, which is builtlocally on the AMPs. The size of Spool 1 is estimated to be 24 rows. Theestimated time for this step is 0.15 seconds.

4) Finally, we send out an END TRANSACTION step to all AMPs involved inprocessing the request.

-> The contents of Spool 1 are sent back to the user as the result of statement 1. Thetotal estimated time is 0 hours and 0.15 seconds.


EXPLAIN Facility – Teradata SQL Extension

The EXPLAIN modifier in front of any SQL statement generates an English translation ofthe Parser’s plan.

The request is fully parsed and optimized, but not actually executed.

EXPLAIN returns:• Text showing how a statement will be processed (a plan)• An estimate of how many rows will be involved• A relative cost of the request (in units of time)

This information is useful for:• predicting row counts• predicting performance• testing queries before production• analyzing various approaches to a problem EXPLAIN

EXPLAIN SELECT last_name, department_number FROM employee ;

Explanation (partial):

3) We do an all-AMPs RETRIEVE step from CUSTOMER_SERVICE.employee by way of an all-rowsscan with no residual conditions into Spool 1, which is built locally on the AMPs. The size ofSpool 1 is estimated to be 24 rows. The estimated time for this step is 0.15 seconds.



Teradata Features ReviewThe Teradata system is a high-performance database system which permits the processing ofenormous quantities of detail data, quantities which are beyond the capability ofconventional systems.

The system is specifically designed for large relational databases. From the beginningthe Teradata system was created to do one thing: manage enormous amounts of data.

Up to one hundred terabytes of on-line storage capacity is currently available making itan ideal solution for enterprise data warehouses or even smaller data marts.

Uniform data distribution across multiple processors facilitates parallel processing. Thesystem is designed in such a way that the component parts divides the work up intoapproximately equal pieces. This keeps all the parts busy all the time; this enables thesystem to accommodate a larger number of users and/or more data.

Open architecture adapts readily to new technology. As higher-performance industrystandard computer chips and disk drives are made available, they are easily incorporatedinto the architecture.

As the configuration grows, performance increase is linear. There is no "wall" ofperformance drop-off until extremely large configurations are attempted.

Structured Query Language (SQL) is the industry standard for communicating withrelational databases.

The Teradata RDBMS currently runs as a database server on a variety of Unix basedhardware platforms, with Windows NT the most recent platform addition.


Teradata Features Review

• Designed for large decision-support queries• Ideal for data warehouse applications• Parallelism makes possible access to very large tables• Performance increase is linear as components are added• Uses standard SQL• Runs as a “database server” to client applications• Runs on multiple hardware platforms• Open architecture - uses industry standard components

TeradataDATABASE

Win 2000

IBMMainframe

UNIX

Win XP


Review Questions

1. What language is used to access a Teradata table?

2. What are five Teradata database objects?

3. What are four major components of the Teradata architecture?

4. What are views?

5. What are macros?



Notes

Teradata RDBMS Architecture Page 4-1

Module 4


Teradata RDBMS Architecture


• Describe the purpose of the PE and the AMP.

• Describe the overall RDBMS parallel architecture.

• Describe the relationship of the RDBMS to its clientside applications.


Page 4-2 Teradata RDBMS Architecture

Notes


Table of Contents

TERADATA AND MPP SYSTEMS ...................................................................................................................... 4-4TERADATA FUNCTIONAL OVERVIEW.......................................................................................................... 4-6CHANNEL-ATTACHED CLIENT SOFTWARE OVERVIEW ........................................................................ 4-8NETWORK-ATTACHED CLIENT SOFTWARE OVERVIEW..................................................................... 4-10THE PARSING ENGINE ..................................................................................................................................... 4-12MESSAGE PASSING LAYER............................................................................................................................. 4-14THE ACCESS MODULE PROCESSOR (AMP) ............................................................................................... 4-16TERADATA PARALLELISM............................................................................................................................. 4-18REVIEW QUESTIONS......................................................................................................................................... 4-20



Teradata and MPP SystemsTeradata is the software that makes a MPP system appear to be a single system to users andadministrators.

The BYNET (BanYan NETwork) is the software and hardware interconnect that provideshigh performance networking capabilities to NCR MPP (Massively Parallel Processing)systems.

Using communication switching techniques, the BYNET allows for point-to-point, multi-cast, and broadcast communications among the nodes, thus supporting a monumentalincrease in throughput in very large databases. This technology allows Teradata users togrow massively parallel databases without fear of a communications bottleneck for anydatabase operations.

Although the BYNET software also supports the multicast protocol, Teradata has notutilized this protocol until V2R5. Teradata software will use the point-to-point protocolwhenever possible. When an all-AMP operation is needed, Teradata software uses thebroadcast protocol to broadcast the request to the AMPs.

The BYNET is linearly scalable for point-to-point communications. For each new nodeadded to the system, an additional 10 MB (or 60 MB with BYNET Version 2) of additionalbandwidth is added to each BYNET, thus providing scalability as the system grows.Scalability comes from the fact that multiple point-to-point circuits can be establishedconcurrently. With the addition of another node, more circuits can be establishedconcurrently.

For broadcast and multicast operations, bandwidth is fixed at 10 MB per second (or 60 MBwith BYNET Version 2) per BYNET.


Teradata and MPP Systems

Teradata is the software that makes a MPP system appear tobe a single system to users and administrators.

BYNET 0 BYNET 1

Node 0

PEPE

AMP AMP

AMP AMP

AMP AMP

AMP AMP

O.S.

PEPE

AMP AMP

AMP AMP

AMP AMP

AMP AMP

O.S.

PEPE

AMP AMP

AMP AMP

AMP AMP

AMP AMP

O.S.

PEPE

AMP AMP

AMP AMP

AMP AMP

AMP AMP

O.S.

Node 1 Node 2 Node 3

The two major componentsof the Teradata RBDMS areimplemented as virtualprocessors (vproc).

• Parsing Engine (PE)• Access Module

Processor (AMP)

The MPL (PDE and BYNET)connects multiple nodestogether and allowsTeradata to communicatebetween nodes.

Administrator

Sticky Note

Massively Parallel Processing

Administrator

Sticky Note

Parallel Database Extension: Software runs on Unix to support parallel environment



Teradata Functional OverviewThe client may be a mainframe system, such as IBM or Unisys in which case it is channel-attached to the Teradata RDBMS. Also, a client may be a PC or UNIX-based system that isLAN or network-attached.

The client application submits an SQL request to the RDBMS, receives the response, andsubmits the response to the user.


Teradata Functional Overview

Teradata RDBMS


Channel-Attached System

LAN

Network-Attached System

ParsingEngine

ParsingEngine

AMP

ClientApplication

CLI or ODBC

MTDP

MOSI

ClientApplication

CLI

TDP

AMP AMP AMP

Channel

Administrator

Sticky Note

Call Level Interface: Creates session between user and DB, create and de-block Parcel of Informations, buffer allocation

Administrator

Sticky Note

Teradata Director Program: Exisit only in Channel Attached, used to balance session, and physical input to PE



Channel-Attached Client Software OverviewIn channel-attached systems, there are three major software components, which playimportant roles in getting the requests to and from the Teradata RDBMS.

The client application is either written by a programmer or is one of Teradata’s providedutility programs. Many client applications are written as “front ends” for SQL submission,but they also are written for file maintenance and report generation. Any client-supportedlanguage may be used provided it can interface to the Call Level Interface (CLI).

For example, a user could write a COBOL application with “embedded SQL”. Theapplication developer would have to use the Teradata COBOL Preprocessor and COBOLcompiler programs to generate an object module and link this object module with the CLI.The CLI application interface provides maximum control over Teradata connectivity andaccess.

The Call Level Interface (CLI) is the lowest level interface to the Teradata RDBMS. Itconsists of system calls which create sessions, allocate request and response buffers, createand de-block “parcels” of information, and fetch response information to the requestingclient.

The Teradata Director Program (TDP) is a Teradata-supplied program that must run onany client system that will be channel-attached to the Teradata RDBMS. The TDP managesthe session traffic between the Call-Level Interface and the RDBMS. Its functions includesession initiation and termination, logging, verification, recovery, and restart, as well asphysical input to and output from the PEs, (including session balancing) and themaintenance of queues. The TDP may also handle system security.

The Host Channel Adapter is a mainframe hardware component that allows the mainframeto connect to an ESCON or Bus/Tag channel.

The PBSA (PCI Bus ESCON Adapter) is a PCI adapter card that allows a WorldMark serverto connect to an ESCON channel.

The PBCA (PCI Bus Channel Adapter) is a PCI adapter card that allows a WorldMarkserver to connect to a Bus/Tag channel.


Channel-Attached Client Software Overview

Client Application- Your own application(s)- Teradata utilities (BTEQ, etc.)

CLI (Call-Level Interface) Service Routines- Request and Response Control- Parcel creation and blocking/unblocking- Buffer allocation and initialization

TDP (Teradata Director Program) - Session balancing across multiple PEs

- Insures proper message routing to/from RDBMS- Failure notification (application failure, Teradata restart)

Channel (ESCON or Bus/Tag)

Channel-Attached System

TDP

ClientApplication

CLI

ClientApplication

CLI

ParsingEngine

ParsingEngine

Host Channel Adapter PBSA or PBCA

Administrator

Sticky Note

Its a hardware used to connect to ESCON or BUS/TAG channel

Administrator

Sticky Note

PCI adapter that connects to BUS/TAG or ESCON



Network-Attached Client Software OverviewIn network-attached systems, there are four major software components that playimportant roles in getting the requests to and from the Teradata RDBMS.

The client application is written by the programmer using a client-supported language suchas “C”. The purpose of the application is usually to submit SQL statements to the RDBMSand perform processing on the result sets. The application developer can “embed” SQLstatements in the application and use the Teradata Preprocessor to interpret the embeddedSQL statements.

In a networked environment, the application developer can use either the CLI interface orthe ODBC driver to access Teradata.

Note: The CLI application interface provides maximum control over Teradataconnectivity and access. The ODBC driver is a much more open standard and iswidely used with client applications to access an RDBMS.

The Call Level Interface (CLI) is a library of routines that resides on the client side. Clientapplication programs use these routines to perform operations such as logging on and off,submitting SQL queries and receiving responses which contain the answer set. Theseroutines are 98% the same in a network-attached environment as they are in a channel-attached.

The Teradata ODBC™ (Open Database Connectivity) driver uses an open standards-based ODBC interface to provide client applications access to Teradata across LAN-basedenvironments. NCR has ODBC drivers for both UNIX and Windows-based applications.

The Micro Teradata Director Program (MTDP) is a Teradata-supplied program that mustbe linked to any application that will be network-attached to the Teradata RDBMS. TheMTDP performs many of the functions of the channel based TDP including sessionmanagement. The MTDP does not control session balancing across PEs. Connect andAssign Servers that run on the Teradata system handle this activity.

The Micro Operating System Interface (MOSI) is a library of routines providingoperating system independence for clients accessing the RDBMS. By using MOSI, we onlyneed one version of the MTDP to run on all network-attached platforms.


Network-Attached Client Software Overview

CLI (Call Level Interface)- Library of routines for blocking/unblocking requests and responses

to/from the RDBMSODBC™ (Open Database Connectivity) Driver

- Uses open standards-based ODBC interface to provide client applicationsaccess to Teradata

MTDP (Micro Teradata Director Program)- Library of session management routines

MOSI (Micro Operating System Interface) - Library of routines providing OS independent interface

LAN-Attached Servers

LAN (TCP/IP)Client

Application(ex., FastLoad)

CLIMTDPMOSI

ClientApplication(ex., SQLAssistant)

ODBCMTDPMOSI

ParsingEngine

ParsingEngine

Gateway Software (tgtw)

ClientApplication(ex., BTEQ)

CLIMTDPMOSI

Ethernet Adapter



The Parsing EngineParsing Engines (PEs) are made up of the following software components: session control,the Parser, the Optimizer, and the Dispatcher.

Once a valid session has been established, the PE is the component that manages thedialogue between the client application and the RDBMS.

The major functions performed by session control are logon and logoff. Logon takes atextual request for session authorization, verifies it, and returns a yes or no answer. Logoffterminates any ongoing activity and deletes the session’s context. When connected to anEBCDIC host the PE converts incoming data to the internal 8-bit ASCII used by theTeradata RDBMS, thus allowing input values to be properly evaluated against the databasedata.

When a PE receives an SQL request from a client application, the Parser interprets thestatement, checks it for proper SQL syntax and evaluates it semantically. The PE also mustconsult the Data Dictionary/Directory to ensure that all objects and columns exist and thatthe user has authority to access these objects.

The Optimizer’s role is to develop the least expensive plan to return the requested responseset. Processing alternatives are evaluated and the fastest alternative is chosen. Thisalternative is converted to executable steps, to be performed by the AMPs, which are thenpassed to the dispatcher.

The Dispatcher controls the sequence in which the steps are executed and passes the stepson to the Message Passing Layer. It is composed of execution control and response controltasks. Execution control receives the step definitions from the Parser, transmits the stepdefinitions to the appropriate AMP or AMPs for processing, receives status reports from theAMPs as they process the steps, and passes the results on to response control once the AMPshave completed processing. Response control returns the results to the user. The Dispatchersees that all AMPs have finished a step before the next step is dispatched.

Depending on the nature of the SQL request, the step will be sent to one AMP, a few AMPs,or all AMPs.


The Parsing Engine

The Parsing Engine is responsible for:• Managing individual sessions (up

to 120)• Parsing and Optimizing your SQL

requests• Dispatching the optimized plan to

the AMPs• Input conversion (EBCDIC / ASCII)

- if necessary• Sending the answer set response

back to the requesting client

Answer Set Response

ParsingEngine

SQL Request

Parser

Optimizer

Dispatcher


AMP AMP AMP AMP



Message Passing LayerThe Message Passing Layer (MPL) handles the internal communication of the TeradataRDBMS. All communication between PEs and AMPs is done via the Message PassingLayer.

When the PE dispatches the steps for the AMPs to perform, they are dispatched onto theMPL. The messages are routed to the appropriate AMP(s) where results sets and statusinformation are generated. This response information is also routed back to the requestingPE via the MPL.

The Message Passing Layer is a combination of the Teradata PDE software, the BYNETsoftware, and the BYNET interconnect itself.

Depending on the nature of the dispatch request, the communication may be a:

Broadcast - message is routed to all AMPs and PEs on the systemMulti-Cast - message is routed to a group of AMPs Point-to-Point - message is routed to one specific AMP or PE on the system

The technology of the MPL is what makes possible the parallelism of the Teradata RDBMS.

The MPL is implemented differently on different platforms:

PDE and BYNET software - used for multi-node MPP systems and single-node SMPsystems (V2R3 and later). With a single-node SMP, the BYNET device driver isused in conjunction with the PDE even though a physical BYNET network is notpresent.

PDE and VNET - used for single-node SMP systems (Release V2R1.1 and ReleaseV2R2).

YNET - used for proprietary Version 1 systems (DBC/1012 and 3600).

The VNET and YNET implementations are only briefly discussed in this course.



Answer Set Response

ParsingEngine

SQL Request

Message Passing Layer(PDE and BYNET)

AMP AMP AMP AMP

The Message Passing Layer isresponsible for:

• Carrying messages between the AMPsand PEs

• Point-to-Point, Multi-Cast, andBroadcast communications

• Merging answer sets back to the PE• Making Teradata parallelism possible

The Message Passing Layer is acombination of:

• Parallel Database Extensions (PDE)Software

• BYNET Software• BYNET Hardware for MPP systems



The Access Module Processor (AMP)The Access Module Processor (AMP) is responsible for managing a portion of thedatabase. An AMP will control some portion of each table on the system. AMPs do all ofthe physical work associated with generating an answer set including, sorting, aggregating,formatting and converting.

An AMP responds to Parser/Optimizer steps transmitted across the MPL by selecting datafrom or storing data to its disks. For some requests the AMPs may also redistribute a copyof the data to other AMPs.

The Database Manager subsystem resides on each AMP. It receives the steps from theDispatcher and processes the steps. To do that it has the ability to lock databases and tables,to create, modify, or delete definitions of tables, to insert, delete, or modify rows withinthe tables, and to retrieve information from definitions and tables. It collects accountingstatistics, recording accesses by session so those users can be billed appropriately. Finally,the Database manager returns responses to the Dispatcher.

Earlier in this course we discussed the logical organization of data into tables. TheDatabase Manager provides a bridge between that logical organization and the physicalorganization of the data on disks. The Database Manager performs a space managementfunction that controls the use and allocation of space.

AMPs also perform output data conversion, checking the session and changing theinternal, 8-bit ASCII used by Teradata to the format of the requester. This is the reverse ofthe process performed by the PE when it converts the incoming data into internal ASCII.


The Access Module Processor (AMP)

Answer Set Response

ParsingEngine

SQL Request


AMP AMP AMP AMP

AMPs store and retrieve rows to and from disk

The AMPs are responsible for:• Finding the rows requested• Lock management• Sorting rows• Aggregating columns• Join processing• Output conversion and formatting• Creating answer set for client• Disk space management• Accounting• Special utility protocols• Recovery processing



Teradata ParallelismParallelism is at the very heart of the Teradata RDBMS. There is virtually no part of thesystem where parallelism has not been built in. Without the parallelism of the system,managing enormous amounts of data would either not be possible or, best case, would beprohibitively expensive and inefficient.

Each PE can support up to 120 user sessions in parallel. This could be 120 distinct users, ora single user harnessing the power of all 120 sessions for a single application.

Each session may handle multiple requests concurrently. While only one request at a timemay be active on behalf of a session, the session itself can manage the activities of 16requests and their associated answer sets.

The Message Passing Layer was designed such that it can never be a bottleneck for thesystem. Because the MPL is implemented differently for different platforms, this means thatit will always be well within the needed bandwidth for each particular platform’s maximumthroughput.

Each AMP can perform up to 80 tasks in parallel. This means that AMPs are not dedicatedat any moment in time to the servicing of only one request, but rather are multi-threadingmultiple requests concurrently.

Because AMPs are designed to operate on only one portion of the database, they mustoperate in parallel to accomplish their intended results.

In addition to this, the optimizer may direct the AMPs to perform certain steps in parallel ifthere are no contingencies between the steps. This means that an AMP might beconcurrently performing more than one step on behalf of the same request.

A recently added feature called Parallel CLI allows for parallelizing the client application,particularly useful for multi-session applications. This is accomplished by setting a fewenvironmental variables and requires no changes to the application code.

In truth, parallelism is built into the Teradata RDBMS from the ground up!


Teradata Parallelism

Notes:• Each PE can handle up to 120 sessions in parallel.• Each Session can handle multiple REQUESTS.• The Message Passing Layer can handle all message activity in parallel.• Each AMP can perform up to 80 tasks in parallel.• All AMPs can work together in parallel to service any request.• Each AMP can work on several requests in parallel.

Parallelism is built intoTeradata from the

ground up!

Session A

Session B

Session C

Session D

Session E

Session F

PE PEPE

Task 1Task 2Task 3

Task 7Task 8Task 9

Task 4Task 5Task 6

Task 10Task 11Task 12

AMP 1 AMP 4AMP 3AMP 2



Review Questions

1. What are the two software elements that accompany an application on all client side environments?

2. What is the purpose of the PE?

3. What is the purpose of the AMP?

4. How many sessions can a PE support?

Match Quiz ____ 1. CLI ____ 2. MTDP ____ 3. MOSI ____ 4. Parser ____ 5. AMP ____ 6. Message Passing Layer ____ 7. TDP ____ 8. Optimizer ____ 9. Dispatcher ____10. Parallelism

a. Does Aggregating and Lockingb. Validates SQL syntaxc. Connects AMPs and PEsd. Balances sessions across PEse. Provides Client side OS independencef. Library of Session Management Routinesg. PE S/W turns SQL into AMP stepsh. PE S/W sends plan steps to AMPi. Library of Teradata Service Routinesj. Foundation of Teradata architecture



Notes

Teradata System Architecture Page 5-1

Module 5


Teradata System Architecture


• Identify characteristics of various components.

• List two non-NCR platforms that support Teradata.


Page 5-2 Teradata System Architecture

Notes


Table of Contents

TERADATA VERSION 1 PLATFORMS............................................................................................................. 5-4DBC/1012 ARCHITECTURE.................................................................................................................................... 5-4

Interface Processor (IFP)................................................................................................................................. 5-4Communications Processor (COP) .................................................................................................................. 5-4

NCR 3600 ARCHITECTURE.................................................................................................................................... 5-4TERADATA VERSION 2 ARCHITECTURE ..................................................................................................... 5-6COMPARING TERADATA V1 AND V2 ............................................................................................................. 5-8

BENEFITS OF VIRTUAL PROCESSORS...................................................................................................................... 5-8TERADATA VERSION 2 SINGLE NODE (SMP) ............................................................................................ 5-10

VNET (OLDER COMPONENT) ................................................................................................................................ 5-10MULTI-NODE MPP SYSTEM ............................................................................................................................ 5-12EXAMPLE OF 4 NODE TERADATA SYSTEM............................................................................................... 5-14TERADATA CLIQUES ........................................................................................................................................ 5-16BYNET (FOR MPP) .............................................................................................................................................. 5-18BYNET COMMUNICATION PROTOCOLS .................................................................................................... 5-20VPROC INTER-PROCESS COMMUNICATION ............................................................................................ 5-22EXAMPLES OF TERADATA VERSION 2 SYSTEMS.................................................................................... 5-24WHAT MAKES NCR’S MPP PLATFORMS SPECIAL? ................................................................................ 5-26NCR RACK-BASED CABINETS ........................................................................................................................ 5-28SUMMARY ............................................................................................................................................................ 5-30REVIEW EXERCISES ......................................................................................................................................... 5-32



Teradata Version 1 PlatformsTeradata Version 1 platforms have been in existence since 1984, first with the originalDatabase Computer, DBC/1012 and in 1991 with the 3600. Both of these systems are oldertechnologies, however they are still in use at some customer sites.

The Teradata Version 1 RDBMS runs on a proprietary 16-bit operating system know asTOS (Teradata Operating System). All AMPs and PE’s are dedicated hardware processorsconnected using a message-passing layer known as the Ynet.

Both platforms support channel-attached and LAN-attached host systems.

DBC/1012 ArchitectureThe DBC/1012 is a dedicated relational database management system. It consists of thefollowing subsystems:

Interface Processor (IFP)The IFP is the Parsing Engine of the DBC/1012 for channel-attached systems. It translatesrequests from a channel-attached client into internal commands, forwards the commandsover the Ynet to the AMPs, and coordinates the responses as they return over the Ynet fromthe AMPs. An IFP provides the session management, SQL translation and optimization, andAMP response coordination. Multiple sessions are handled by the IFP’s multi-taskingcapability. Multiple IFPs can be used to process additional sessions and increasethroughput.

Communications Processor (COP)The COP is the Parsing Engine of the DBC/1012 for network-attached systems. A COPreceives requests containing Teradata SQL statements originating from network attachedclients, such as a workstation on a Local Area Network (LAN). Its function is the same asthat of the IFP on channel attached clients. The COP manages the network traffic betweenthe client and the DBC/1012. The COP translates requests from the client into internalcommands, forwards the commands over the Ynet to the AMPs, and then coordinates theresponses as they return.

NCR 3600 ArchitectureThe NCR 3600 system is a multipurpose, UNIX-based Massively Parallel System (MPP).It can run all UNIX-based application and can also run the Teradata RDBMS under TOS.UNIX applications are run on AP’s while Teradata is run on PEs and AMPs. All processingunits are connected via the Ynet.


Teradata Version 1 Platforms

Teradata Version 1 was a combination of hardware and software.

For example, if a customer needed additional Parsing Engine capability, thehardware and software components for a “Parsing Engine” had to bepurchased, installed, and configured.

Platform Year Available Upper Limit CPUs

Teradata DBC/1012 1984 1 TB 286, 386, 486

NCR 3600 1991 4 TB 486

Both platforms have the following common characteristics:TOS based (Teradata Operating System)Hardware PE Referred to as IFP or COP on DBC/1012Hardware AMP Intel 286, 386, or 486 CPUsYnet Message Passing LayerChannel-Attached Hosts Bus and Tag onlyLAN Attached Hosts

This course will not discuss Version 1 platforms in any detail.

Administrator

Sticky Note

Interface Processor: translate commands from channel attached to internal commands

Administrator

Sticky Note

Communication Processor: Translate requests from network clients to internal commands



Teradata Version 2 ArchitectureWith Teradata Version 2 (available to customers in January 1996), the Teradata RDBMSbecame an open database system. No longer was Teradata software dependent on aproprietary Operating System (TOS) and proprietary hardware. Rather, it was anapplication that ran under UNIX. UNIX MP-RAS is the NCR version of UNIX SVR5 formulti-processor systems. Teradata RDBMS V2R5.0 is available as a database with UNIXMP-RAS and Windows 2000.

TOS was replaced with 32-bit Parallel Database Extensions (PDE). PDE software providesa TOS-like functionality for the Teradata RDBMS and allows Teradata to run under UNIXMP-RAS.

Another significant difference was the ability to run Teradata on a UNIX based platformwith “virtual AMPs” and “virtual PEs”, rather than the hardware-based AMP and PEs ofVersion 1.

By porting Teradata to a general-purpose UNIX platform, a variety of processing optionsagainst the Teradata database became possible, all within a single system. OLTP, as well asOLCP and OLAP applications, became processing options in addition to standard DSS.

Because of the scalability of the MPP system, it became possible to purchase a small, singlenode SMP system for databases as small as 10 GB and to eventually grow incrementally to amultiple terabyte system. Low cost development systems were also an additional benefit tothe SMP platform.

Significant performance advantages over Version 1 became possible with the running of 32-bit PDE as a replacement to 16-bit TOS. Additionally, only a single copy of PDE need runper node, as opposed to TOS having to run on each physical processor.

Version 2 also added substantial new features and functionality to the product, including anew set of installation and operational tools. Also, Version 2 made possible an upgrade ofSQL to conform to ANSI standards along with new added features of the language.

An application, which runs under the control of PDE, such as the Teradata RDBMS, isconsidered a Trusted Parallel Application (TPA). The Teradata RDBMS is the only TPAapplication available at this time.


Teradata Version 2 Architecture

PE vproc

AMPvproc

Vdisk

PE vproc

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

PDE and BYNET S/W (Message Passing Layer)

Operating System (UNIX MP-RAS or Windows 2000)

• Teradata RDBMS V2R5 executes on UNIX MP-RAS or Windows 2000.• Uses general purpose SMP/MPP hardware.• Virtual AMPs and PEs (Vprocs) replace hardware AMPs and PEs.• “Shared Nothing” Architecture - each AMP has its own memory and manages its own

disk space.• Parallel Database Extensions (32 bit) replaces TOS (16 bit). PDE is unique per OS.• Can run DSS, OLTP, OLCP, or OLAP applications.• Major performance and feature enhancements over V1.

Administrator

Sticky Note

Online Complex Processing: involves dynamic arrays, large objects, multimedia, or other complex data types



Comparing Teradata V1 and V2The large change from Teradata V1 to V2 was the implementation of “Virtual Processors.This allowed Teradata to go from a hardware/software implementation to a softwareimplementation.

Virtual processors, (vprocs), represent a significant change in the implementation of theTeradata RDBMS under V2. Using UNIX (PDE), Vprocs replace the physical processors(IFPs, COPs, and AMPs) found in the DBC and 3600. A vproc is a collection of processeswithin the SMP node. Vprocs are defined in two ways: logically and physically.

Logically, a vproc is considered a separate instance of an AMP or PE within an SMP node.More than one logical AMP or PE can run on an SMP node. A vproc functions the same asa physical AMP or PE did on a 3600.

Physically, a vproc is an addressable collection of processes that can share resources.

Each vproc is a separate task running under the multi-tasking environment of the UNIXoperating system. It is a separate copy of the processor software, isolated from the othervprocs.

Benefits of Virtual ProcessorsVprocs offer a number of advantages over physical processors.

Applications other than the Teradata DBS can execute on the same node. Examples includeapplications and Teradata utilities. However, to optimize Teradata performance, the nodeshould be dedicated to Teradata.

Database software resets no longer cause a hardware and operating system reboot. UNIXresets the task.

All SMP benefits are now available for the Teradata DBS. For example, a memory pool isshared among the vprocs; in-memory message passing permits vproc communication, and soon. Less memory is required for the operating system. A single copy of UNIX serves theentire node. Compare that to TOS, which must run a separate copy on each AMP and PE.

There is a better balance of processing power and disks and a higher utilization of the SMP.

Finally, the number of AMPs is variable and can be configured depending on the workload.


Comparing Teradata V1 and V2

AMPTOS

AMPTOS

AMPTOS

PETOS

AMPTOS….

Operating System (e.g., UNIX MP-RAS)PDE

Vdisk Vdisk Vdisk Vdisk Vdisk Vdisk Vdisk Vdisk Vdisk Vdisk

PE vproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

PE vproc

Version 1 CharacteristicsPEs and AMPs are physical processors.

AMPs map to physical or logical disks.

Each AMP runs TOS (16-bit OS).

Each PE/AMP only has 16 MB of memory.

Fastest CPU utilized was Intel i486 CPU.

Version 2 CharacteristicsPEs and AMPs are virtual processors.

AMPs map to virtual disks (Vdisks).

Each SMP node runs and OS and PDE.

Each SMP can have up to 4 GB of physicalmemory.

Utilizes newest CPUs (e.g., Intel PentiumIV Xeon 3.06 GHz).



Teradata Version 2 Single Node (SMP)Under the single-node (SMP) Version 2 of Teradata, all applications run under UNIX andall Teradata software runs under PDE. All share the resources of CPUs and memory of theSMP node

In addition to user applications, LAN gateway software also runs as UNIX processes. Achannel driver is added as part of UNIX.

AMPs and PEs are no longer instances of hardware processors, but are now “virtualprocessors - vprocs”. They run under the control of PDE and their number is softwareconfigurable.

AMPs are associated with “virtual disks – vdisks” which are associated with logical units(LUNs) within a disk array.

For single-node UNIX implementations with Teradata Version 2 (Releases 1 and 2), theAMPs and PEs communicated with each other via the “virtual network – Vnet”. Thisinternal network is simply a memory based message flow control system. PDE actuallycontrols the message passing activity, while the Vnet handles the message queuing and flowcontrol.

Vnet (Older component)The Vnet is a virtual network that replaces the Ynet in single-node SMP systems forTeradata V2R1 and V2R2 configurations. The Vnet used shared memory to handle queuingof messages between AMP vprocs and PE vprocs within the same SMP node. The VnetInterface controls the flow of messages over the Vnet to and from the vprocs.

The Vnet doesn't really move data when it queues messages or redistributed rows. Instead,it changes the owner of the memory location to that of the destination vproc. Memorylocations are assigned to one vproc at a time – they never share memory locations.

The PDE software actually controls the message passing activity, while the Vnet handles themessage queuing and flow control.

Under V2R3 of Teradata, the Vnet is replaced by BYNET software (a.k.a., “boardlessBYNET”) for single SMP systems. More details on the BYNET are provided later in thiscourse.


Teradata Version 2 Single Node (SMP)

CHANNEL LAN

PC W/SChannel-Attached Systems

TDP

Applications/Utilities

Channel Drivers Gateway Software

PE vproc

AMPvproc

Vdisk

PE vproc

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

AMPvproc

Vdisk

PDE and BYNET S/W (Message Passing Layer)

Operating System (UNIX MP-RAS or Windows 2000)

Notes:• Applications, channel

drivers, and the LANgateway software run asOperating Systemprocesses.

• AMPs and PEs are virtualprocessors (Vprocs) whichrun under the ParallelDatabase Extensions (PDE).

• AMPs are associated withvirtual disks (Vdisks) withina disk array subsystem.

• Teradata is called a TrustedParallel Application of PDE.

Administrator

Sticky Note

SMP: Symmetric Multi Processing, it is a single node



Multi-Node MPP SystemWhen multiple SMP nodes are connected together to form a larger configuration, we refer tothis as an MPP (Massively Parallel Processing) system.

The connecting layer (or system interconnect) is called the BYNET. The BYNET is acombination of hardware and software that allows multiple vprocs on multiple nodes tocommunicate with each other.

Because Teradata is a linearly expandable database system, as additional nodes and vprocsare added to the system, the system capacity scales in a linear fashion.

The BYNET Version 1 can support up to 128 SMP nodes. The BYNET Version 2interconnect can support up to 512 SMP nodes.

Acronyms that may appear in diagrams in this module:

PCI – Peripheral Component InterconnectEISA – Extended Industry Standard ArchitecturePBCA – PCI Bus Channel AdapterPBSA – PCI Bus ESCON AdapterEBCA – EISA Bus Channel Adapter


Multi-Node MPP System

BYNET

DAC DAC DAC DAC DAC DAC DAC DAC

SMP SMP SMP SMP

Notes:• Multiple nodes may be

configured to provide aMassively ParallelProcessing (MPP) system.

• A physical message passinglayer called the BYNET isneeded to interconnectmultiple nodes.

• Teradata is a linearlyexpandable RDBMS - asyour database grows,additional nodes may beadded.

Administrator

Sticky Note

MPP: Massively Parallel Processing, it consist of multi nodes



Example of 4 Node Teradata SystemThe facing page contains an illustration of a simple four-node Teradata system. Eachnode has its own Vprocs to manage, while communication among the Vprocs takes place viathe BYNETs. The PEs are not shown in this example.

Each node is an SMP from a configuration standpoint. Each node has its own CPUs,memory, UNIX and PDE software, Teradata RDBMS software, BYNET software, andaccess to one or more disk arrays.

Nodes are the building blocks of MPP systems. A system size is typically expressed interms of number of nodes.

AMPs provide access to user data stored within tables that are physically stored on diskarrays.

Each AMP is associated with a Vdisk. Each AMP sees its Vdisk as a single disk. Teradata(AMP software) organizes its data on its disk space (Vdisk) using a Teradata “File System”structure.

A Vdisk may be actually composed of multiple Pdisks - Physical disk. A Pdisk is assignedto physical drives in a disk array.


Example of 4 Node Teradata System

DAC-A DAC-BDAC-A DAC-BDAC-A DAC-B DAC-A DAC-B

0 4 36…….

SMP001-4 AMPs

1 5 37…….

SMP001-5 AMPs

2 6 38…….

SMP002-4 AMPs

3 7 39…….

SMP002-5 AMPs

BYNET

RAID 1

Vdisk 0

Pdisk 0

Pdisk 1

MaxPerm 72 GB

AMP 0

36 GB

36 GB

36 GB

36 GB

Administrator

Sticky Note

DAC: Discretionary Access Control



Teradata CliquesA clique is a set of Teradata nodes that share a common set of disk arrays. In the event ofnode failure, all vprocs can migrate to another available node in the clique. All nodes in theclique must have access to the same disk arrays.

The illustration on the facing page shows an 8-node system consisting of two cliques, eachcontaining four nodes. Because all disk arrays are available to all nodes in the clique, theAMP vprocs will still have access to the rows they are responsible for.

In the event of three out of four nodes failing, the remaining node would attempt to absorball vprocs from the failed nodes. Because each node can support a maximum of 128 vprocs,the total number of vprocs for the clique should not exceed 128.


Teradata Cliques

• A clique is a defined set of nodes that share a common set of disk arrays.• All nodes in a clique must be able to access all Vdisks for all AMPs in the clique.• A clique provides protection from a node failure.• If a node fails, all vprocs will migrate to the remaining nodes in the clique (Vproc

Migration).• A clique can support up to 128 vprocs.

BYNET

SMP SMP SMP SMP SMP SMP SMP SMP

Clique 0 Clique 1



BYNET (for MPP)There are two physical BYNETs, BYNET 0 and BYNET 1. Both are fully operational andprovide fault tolerance in the event of a BYNET failure. The BYNETs automatically handleload balancing and message routing. BYNET reconfiguration and message rerouting in theevent of a component failure is also handled transparently to the application.

A future possibility of BYNET Version 2 will be to have 4 BYNETs in a single system.


BYNET (for MPP)

BYNET 0 BYNET 1

SMP SMP SMP SMP SMP SMP SMP SMP

The BYNET is a dual redundant, bi-directional interconnect network.

All SMPs are connected to both BYNETs.

BYNET Features:

• Enables multiple SMP nodes to communicate.• Automatic load balancing of message traffic.• Automatic reconfiguration after fault detection.• Fully operational dual BYNETs provide fault tolerance.• Scalable bandwidth as nodes are added.• Even though there are two physical BYNETs to provide redundancy and

bandwidth, Teradata and TCP/IP software only see a single network.



BYNET Communication ProtocolsUsing communication-switching techniques, the BYNET allows for point-to-point,multicast, and broadcast communications among the nodes, thus supporting a monumentalincrease in throughput in very large databases. This technology allows Teradata users togrow massively parallel databases without fear of a communications bottleneck for anydatabase operations.

Although the BYNET software supports the multi-cast protocol, Teradata only uses thisprotocol with Group AMPs operations, which is a Teradata V2R5 feature. Teradatasoftware will use the point-to-point protocol whenever possible. When an all-AMPoperation is needed, Teradata software uses the broadcast protocol to send messages to thedifferent SMPs.

Unlike the Ynet, the BYNET is linearly scalable for point-to-point communications. Foreach new node added to the system, an additional 10 MB (or 60 MB with BYNET Version2) of additional bandwidth is added to each BYNET, thus providing scalability as the systemgrows. Scalability comes from the fact that multiple point-to-point circuits can beestablished concurrently. With the addition of another node, more circuits can beestablished concurrently.

For broadcast and multicast operations, bandwidth is fixed at 10 MB per second (or 60 MBwith BYNET Version 2) per BYNET.


BYNET Communication Protocols

BYNET 0 BYNET 1

PE

AMP ... AMP

PE PE

AMP ... AMP

PE PE

AMP ... AMP

PE PE

AMP ... AMP

PE

Point-to-Point (one-to-one):Any node communicates with one node. Scalable bandwidth:

BYNET V1 - 10 MB x 2 BYNETs = 20 MB per nodeBYNET V2 - 60 MB x 2 BYNETs = 120 MB per node

Multi-Cast (one-to-many):Any node communicates to a subset of nodes.Used with Group AMP operations - Teradata release V2R5 and above.

Broadcast (one-to-all):Any node communicates to all nodes. Not scalable: BYNET V1 - Fixed at 10 MB per BYNET

BYNET V2 - Fixed at 60 MB per BYNET

Administrator

Sticky Note

2 way, inbound & outbound



Vproc Inter-process Communication The “message passing layer” is a combination of two pieces of software and hardware– thePDE and the BYNET device drivers and software and the BYNET hardware.

Communication among vprocs in an MPP system may be either inter-node or intra-node.When vprocs within the same node communicate they do not require the physical transportservices of the BYNET. They do however, use the highest levels of the BYNET softwareeven though the messages themselves do not leave the node.

When vprocs must communicate across nodes, they must use the physical transport servicesof the BYNET requiring movement of the data. Any broadcast messages, for example, willgo out to the BYNET, even for the AMPs and PEs that are in the same node.

With Teradata for UNIX Version 2 Releases 1 and 2 (V2R1 and V2R2), communicationamong vprocs in a single SMP system occurs via the Vnet. The Vnet is a memory-basedmessage passing layer. One vproc writes something to memory and makes that addressavailable to another vproc who may then read it. There is no physical movement of data.

With Teradata for UNIX Version 2 Release 3 (V2R3) and V2R4, communication amongvprocs in a single SMP system occurs with the PDE and BYNET software, even though aphysical BYNET does not exist in a single-node system.


Vproc Inter-process Communication

Single-Node or SMP System

Teradata V2R3 and above

Teradata RDBMS

PDE and BYNET s/w

vproc vproc vproc

vproc vproc vproc

BYNET

Node B

Teradata RDBMS

PDE and BYNET s/w

Node A

Teradata RDBMS

PDE and BYNET s/w

vproc vproc

vproc vproc

vproc vproc vproc

vproc vproc vproc

vproc

vproc

MPP Systems

Teradata V2R3 and above

Administrator

Sticky Note

BYNET: is the hardware



Examples of Teradata Version 2 Systems The facing page identifies various SMP servers and MPP systems that are supported forTeradata.

The following dates indicate when these systems were generally available to customers(GCA – General Customer Availability).

– 5100M January, 1996 (not described in this course)– 4700/5150 January, 1998 (not described in this course)– 4800/5200 April, 1999– 4850/5250 June, 2000– 4851/4855/5251/5255 July, 2001– 4900/5300 March, 2002– 4950/5350 December, 2002– 4980/5380 August, 2003

Teradata is also available on non-NCR platforms. Teradata (for Windows 2000) is availableon the Intel-based mid-range platforms running Microsoft Windows 2000. Examples ofother servers that support Teradata are:

• HP NetServer (2-way or 4-way SMP)• Compaq ProLiant (2-way or 4-way SMP)• Dell PowerEdge (2-way or 4-way SMP)

Teradata V2R4.1 and future releases are not supported with Windows NT, but are supportedwith Windows 2000.


Examples of Teradata Version 2 Systems

Examples of NCR systems that support Teradata include:SMP Servers 1 – 4 Nodes 2 – 512 NodesNCR 4400 NCR 4800/4850 NCR 5200/5250NCR 4455 NCR 4851/4855 NCR 5251/5255NCR 4470 NCR 4900 NCR 5300NCR 4475 NCR 4950 NCR 5350NCR 4480 NCR 4980 NCR 5380

The basic building block is the SMP (Symmetric Multi-Processing) node.Common characteristics of these systems:

• MPP systems use the BYNET interconnect• Support of the Teradata database – Version 2• Single point of operational control – AWS• Rack-based systems – each technology is encapsulated in its own chassis

Key differences:• Speed and capacity of SMP nodes and systems• Cabinet architecture• BYNET interface cards, switches and speeds

Examples of other SMP vendors (running Windows 2000) are Dell, HP, andCompaq.

Administrator

Sticky Note

Administration Workstation



What makes NCR’s MPP Platforms Special?The facing page lists the major features of NCR’s MPP systems.

Acronyms:PUT – Parallel Upgrade UtilityAWS – Administration Workstation


What Makes NCR’s MPP Platforms Special?

Key features of NCR’s MPP systems include:

• Teradata RDBMS software - allows Teradata to execute on multiple SMPsand act as a single instance.

• Scalable BYNET Interconnect - as you add nodes, you add bandwidth.

• PUT - Parallel Upgrade Utility - simplifies installation/upgrade of softwareacross many SMPs

• AWS - single point of operational control and scalable server management.

• SMP operating system software (e.g., UNIX MP-RAS and Windows 2000) isonly aware of the resources within the SMP and only has to manage thoseresources.

• Dual AC capability for increased availability.



NCR Rack-Based CabinetsThe facing page contains two pictures of rack-based cabinets. The left picture identifies anSMP cabinet housing four 5380 SMP nodes. The right picture identifies a disk array cabinethousing two disk array subsystems.

This same rack or cabinet is also used for the 4700, 4800, 4850, 4851, 4855, 4900, 4950,4980, 5200, 5250, 5251, 5255, 5300, 5350, and 5380 systems. The processor cabinet or rackis similar in appearance and size to the NCR Enterprise Storage cabinets.

Specifications of a this rack or cabinet are:

Height - 77”Width - 25.25”Depth - 40”Load limit - 1,100 pounds (499 kg.)

This industry-standard rack is referred to as a 40U rack where a U is a unit of measure of1.75”.


NCR Rack-Based Cabinets

with door

SMPxxx-4

BYNET V2

BYNET V2

SMC

SMPxxx-5

SMPxxx-6

SMPxxx-7

NCR 5380

SMC - 2U

6841-2456

6841-2456

NCR 6841

NCR uses industrystandard rack-basedcabinets.

Shown are examplesof a Processor rackand a Storage or diskarray rack.



SummaryThe facing page summarizes the key points and concepts discussed in this module.


Summary

• Teradata Version 1 was a combination of hardware and software.

• Teradata RDBMS Version 2 is a software implementation of Teradata.

– Virtual AMPs and PEs (Vprocs) replace hardware AMPs and PEs.

• Teradata utilizes a “Shared Nothing” Architecture - each AMP has its ownmemory and manages its own disk space.

• Teradata is called a Trusted Parallel Application.

• Multiple nodes may be configured to provide a Massively ParallelProcessing (MPP) system.

• A clique is a defined set of nodes that share a common set of disk arrays.

• Teradata is a linearly expandable RDBMS - as your database grows,additional nodes may be added.



Review ExercisesCheck your understanding of the concepts discussed in this module by completing thereview questions as directed by your instructor.


Review Questions

Complete the following.

1. Each AMP has its own memory and manages its own disk space and executes independently ofother AMPs. This is referred to as a ___________ __________ architecture.

2. The software component that allows Teradata to execute in different operating systemenvironments is the __________.

3. A physical message passing interconnect is called the _____________.

4. A clique provides protection from a _________ failure.

5. If a node fails, all vprocs will migrate to the remaining nodes in the clique. This feature is referredto as ___________ _____________.

6. The ____________ provides a single point of operational control for NCR MPP systems.

7. List two non-NCR platforms that Teradata is supported on. _____________ ____________



Notes

Creating a Teradata Database Page 6-1

Module 6


Creating a Teradata Database


• Distinguish between a Teradata Database and Teradata User.

• Define Perm Space and explain how it is used.

• Define Spool Space and its use.

• Visualize the hierarchy of objects in a Teradata system.


Page 6-2 Creating a Teradata Database

Notes


Table of Contents

A TERADATA DATABASE .................................................................................................................................. 6-4Tables ............................................................................................................................................................... 6-4Views ................................................................................................................................................................ 6-4Macros.............................................................................................................................................................. 6-4Triggers ............................................................................................................................................................ 6-4

A TERADATA USER ............................................................................................................................................. 6-6DATABASE – USER COMPARISON .................................................................................................................. 6-8THE HIERARCHY OF DATABASES AND USERS ........................................................................................ 6-10EXAMPLE OF A SYSTEM HIERARCHY........................................................................................................ 6-12PERMANENT SPACE.......................................................................................................................................... 6-14SPOOL SPACE...................................................................................................................................................... 6-16SYSTEM VARIABLES AND KEYWORDS....................................................................................................... 6-18CREATING TABLES ........................................................................................................................................... 6-20DATA TYPES ........................................................................................................................................................ 6-22ACCESS RIGHTS AND PRIVILEGES .............................................................................................................. 6-24REVIEW QUESTIONS......................................................................................................................................... 6-26



A Teradata DatabaseA Teradata database is a collection of tables, views, macros, triggers, stored procedures,access rights and space limits used for administration and security. All databases have adefined upper limit of permanent space. Permanent space is used for storing the data rowsof tables. Perm space is not pre-allocated. It represents a maximum limit. All databasesalso have an upper limit of spool space. Spool space is temporary space used to holdintermediate query results or formatted answer sets to queries.

Databases provide a logical grouping for information. They are also the foundation forspace allocation and access control. We'll review the definitions of tables, views, andmacros.

TablesA table is the logical structure of data in an RDBMS. It is a two-dimensional structuremade up of columns and rows. A user defines a table by giving it a table name that refersto the type of data that will be stored in the table.

A column represents attributes of the table. Attributes identify, describe, or qualify thetable. Column names are given to each column of the table. All the information in acolumn is the same type, for example, data of birth.

Each occurrence of an entity is stored in the table as a row. Entities are the people, things,or events that the table is about. Thus a row would represent a particular person, thing, orevent.

ViewsA view is a pre-defined subset of one of more tables or other views. It does not exist as areal table, but serves as a reference to existing tables or views. One way to think of a viewis as a virtual table. Views have definitions in the data dictionary, but do not contain anyphysical rows. Views can be used by the database administrator to control access to theunderlying tables. Views can be used to hide columns from users, to insulate applicationsfrom database changes, and to simplify or standardize access techniques.

MacrosA macro is a definition containing one or more SQL commands and report formattingcommands that is stored in the Data Dictionary/Directory. Macros are used to simplify theexecution of frequently-used SQL commands.

TriggersA trigger consists of one or more SQL statements that are associated with a table andare executed when the trigger is “fired”.


A Teradata Database

A Teradata database is a defined logical repository for:• Tables• Views• Macros• Triggers• Stored Procedures

Attributes that may be specified for a database:• Perm Space - maximum amount of space available for tables• Spool Space - maximum amount of work space available for requests• Temp Space - maximum amount of temporary table space

A Teradata database is created with the CREATE DATABASE command.

Example CREATE DATABASE Database_2 FROM SysdbaAS PERMANENT = 20e9, SPOOL = 200e6;

“Database_2” is owned by “Sysdba”.

A database is empty until objects are created within it.



A Teradata UserA user can also be thought of as a collection of tables, views, macros, triggers, storedprocedures, and access rights.

A user is almost the same as a database except that a user can actually log on to the DBS.To accomplish this a user must have a password. A user may or may not have perm space.

Even with no perm space, a user can access other databases depending on the privileges theuser has been granted.

Users are created with the SQL statement CREATE USER.


A Teradata User

A Teradata user is a database with an assigned password.

A Teradata user may logon to Teradata and access objects within:• itself• other databases for which it has access rights

Examples of attributes that may be specified for a user:• Perm Space - maximum amount of space available for tables• Spool Space - maximum amount of work space available for requests• Temp Space - maximum amount of temporary table space

A user is an active repository while a database is a passive repository.

A user is created with the CREATE USER command.

CREATE USER User_C FROM User_AAS PERMANENT = 10e6

,SPOOL = 100e6,TEMPORARY = 80e6,PASSWORD = lucky_day ;

Example

“User_C” is owned by “User_A”.A user is empty until objects are created within it.



Database – User ComparisonIn Teradata, a Database and a User are essentially the same. Database/User names must beunique within the entire system and represent the highest level of qualification in an SQLstatement.

A User represents a logon point within the hierarchy and Access Rights apply only to Users.In many systems, end users do not have Perm space given to them. They are granted rightsto access database(s) containing views and macros, which in turn are granted rights to accessthe corporate production tables.

At any time, another authorized User can change the Spool (workspace) limit assigned to aUser.

Databases may be empty. They may not have any tables, views, macros, or triggers. Theymay or may not have Perm Space allocated. The same is true for Users. The only absoluterequirement is that a User must have a password.

Once Perm Space is assigned, then and only then can tables be put into the database.Macros and views may be added at any time, with or without Perm Space.

Remember that databases and users are both repositories for database objects. The maindifference is the user ability to logon and acquire a session with the RDBMS.

A row exists in DBC.Dbase for each User and Database.


Database – User Comparison

User DatabaseUnique Name Unique NamePassword = ValueDefine and use Perm space Define and use Perm spaceDefine and use Spool space Define Spool spaceDefine and use Temporary space Define Temporary spaceSet Fallback protection default Set Fallback protection default Set Permanent Journal defaults Set Permanent Journal defaultsMultiple Account strings One Account stringEstablish a session with a priorityMay have a startup stringDefault database, dateform, timezone,

and default character setCollation Sequence

• You can only LOGON as a known User to establish a session with Teradata.• Tables and Stored Procedures require Perm Space.• Views, Macros, and Triggers are definitions in the DD/D and require no Perm Space.• A database (or user) with zero Perm Space may have views, macros, and triggers, but

cannot have tables or stored procedures.



The Hierarchy of Databases and UsersAs you define users and databases, a hierarchical relationship among them will evolve.

When you create new objects, you subtract permanent space from the assigned limit of anexisting database or user. A database or user that subtracts space from its own permanentspace to create a new object becomes the immediate owner of that new object.

An “owner” or “parent” is any object above you in the hierarchy. (Note that you can use theterms owner and parent interchangeably.) A “child” is any object below you in thehierarchy. An owner or parent can have many children.

The term “immediate parent” is sometimes used to describe a database or user just aboveyou in the hierarchy.


Hierarchy of Databases and Users

User DBC

User SYSDBA

Database_1User_A User D

Database_3

User_CUser_B

Database_2

Maximum Perm Space - availablebut not yet assignedCurrent Perm Space - containstables or stored proceduresNo Perm SpaceNo Box

• A new database or user must be created from an existing database or user.• All Perm space specifications are subtracted from the immediate owner or parent.• Perm space is a zero sum game - the total of all Perm Space allocations must equal the

total amount of disk space available to Teradata.• Perm space is only used for tables and stored procedures.• Perm space currently unused is available to be used as Spool or Temp space.



Example of a System HierarchyAn example of a system structure for the Teradata database is shown on the facing page.


Example of a System Hierarchy

A User and/or aDatabase may be givenPERM space.

In this example, Markand Tom have noPERM space, butSusan does.

Mark

Customer_Service

SysDBA

DBC

CrashDumps SysAdmin SystemFE

Users may use views and macrosto access the actual tables.

CS_View_Mac

View_1View_2

Macro_1Macro_2

CS_Tables

Table_1Table_2Table_3Table_4

Sys_Calendar

Susan

QCD

CS_Users

Tom



Permanent SpacePermanent Space (Perm space) is the maximum amount of storage assigned to a user ordatabase for holding table rows, Fallback tables, secondary index subtables, storedprocedures, and permanent journals.

Perm space is specified in the CREATE statement as illustrated below. Perm space is notpre-allocated which means that it is available on demand, as entities are created not reservedahead of time. Perm space is deducted from the owner’s specified Perm space and isdivided equally among the AMPs. Perm space can be dynamically modified.

The total amount of Perm space assigned divided by the number of AMPs equals the per-AMP limit. Whenever the per AMP limit is exceeded on any AMP, a Database Fullmessage is generated.

CREATE DATABASE CS_Tables FROM Customer_Service ASPERMANENT = 80000000000 BYTES, …


Permanent Space

CREATE DATABASE CS_tables FROM Customer_ServiceAS PERMANENT = 80e9 BYTES, ...

10 GB

AMP

10 GB

AMP

10 GB

AMP

10 GB

AMP

10 GB

AMP

10 GB

AMP

10 GB

AMP

10 GB

AMP

Perm SpaceLimit per AMP

• Table rows, index subtable rows, and stored procedures row use Perm space.• Fallback protection uses twice the Perm space of No Fallback.• Perm space is deducted from the owner’s database space.• Disk space is not reserved ahead of time, but is available on demand.• Perm space is defined globally for a database.• Perm space can be dynamically modified.• The global limit divided by the number of AMPs is the per/AMP limit.• The per/AMP limit cannot be exceeded.• Good data distribution is crucial to space management.



Spool SpaceSpool Space is temporary space acquired automatically by the system and used for workspace and answer sets for intermediate and final results of Teradata SQL statements (e.g.,SELECT statements generally use Spool space to store the SELECTed data).

A Spool limit is specified in the CREATE statement shown below. This limit cannotexceed the Spool limit of the creating user. However, a single user can create multipledatabases or users, and each can have a Spool limit as large as the Spool limit of that singleuser.

The total amount of Spool space assigned divided by the number of AMPs equals the perAMP limit. Whenever the per-AMP limit is exceeded on any AMP, an Insufficient Spoolmessage is generated to that client.

CREATE USER Susan FROM CS_Users ASPERMANENT = 40000000 BYTES,

SPOOL = 240000000 BYTES,PASSWORD = secret ...


Spool Space

CREATE USER Susan FROM CS_Users ASPERMANENT = 40e6 BYTES, SPOOL = 240e6 BYTES, PASSWORD = secret …

30 MB

AMP

30 MB

AMP

30 MB

AMP

30 MB

AMP

30 MB

AMP

30 MB

AMP

30 MB

AMP

30 MB

AMP

Spool SpaceLimit per AMP

• Spool space is work space acquired automatically by the system forintermediate query results or answer sets.

• SELECT statements generally use Spool space.• Only INSERT, UPDATE, and DELETE statements affect table contents.• The Spool limit cannot exceed the Spool limit of the original owner.• The Spool limit is divided by the number of AMPS in the system, giving a per-

AMP limit that cannot be exceeded.• “Insufficient Spool” errors often result from poorly distributed data or joins on

columns with large numbers of non-unique values.• Keeping Spool rows small and few in number reduces Spool I/O.



System Variables and KeywordsA number of system variables and keywords can be selected and used to verify username,account id, session number, and current database.

The following system variables are available for use with SQL queries.

USER - contains the username of the requesting session/user.SESSION - contains the session-id of the requesting session/user.

DATABASE - contains the current default database of the requestingsession/user.

ACCOUNT - contains the user account info of the requesting session/user.

Each is defined internally as a VARCHAR (30).

These system variables can be especially helpful for queries against the Data Dictionary.

The facing page contains an example of using these keywords.

The Kind column codes represent:

T – TableV – ViewM – MacroG – TriggerI – Join IndexP – Stored ProcedureN – Hash IndexJ – Permanent Journal


System Variables and Keywords

As the requesting user, system variables and the help function can be used todisplay information about your session, objects within a database/user, etc.

USER – username ACCOUNT – account idDATABASE – current database SESSION – session #

Example 1:SELECT USER, ACCOUNT, User Account Session Database

SESSION, DATABASE; SUSAN $M_9038 1000 SUSANDATABASE Order_DB;SELECT USER, ACCOUNT, User Account Session Database

SESSION, DATABASE; SUSAN $M_9038 1000 Order_DB

Example 2:HELP USER Order_DB; Table/View/Macro Name TableKind Comment

customer T ?cust_orders V ?cust_ord_ix I ?orders T ?orders_hash_1 N ?orders_ix_cust_id I ?set_ansidate_on M ?set_integerdate_on M ?



Creating TablesCreation of tables is done via the DDL portion of the SQL command vocabulary. The tabledefinition, once accepted, is stored in the DD/D.

Creating tables requires the definition of at least one column and the assignment of aPrimary Index. Columns are assigned data types, attributes and optionally may be assignedconstraints, such as a range constraint.

Tables, like views and macros, may be dropped when they are no longer needed. Droppinga table both deletes the data from the table and removes the definition of the table from theDD/D.

Secondary indexes may also optionally be assigned at table creation, or may deferred untilafter the table has been built. Secondary indexes may also be dropped, if they are no longerneeded. It is not uncommon to create secondary indexes to assist in the processing of aspecific job sequence, then to delete the index, and its associated overhead, once the job iscomplete.

We will have more to say on indexes in general in future modules.


Creating Tables

Creating a table requires ...– defining columns– assigning of a primary index– optional assignment of secondary indexes

CREATE TABLE Employee(Employee_Number INTEGER NOT NULL,Last_Name CHAR(20) NOT NULL,First_Name VARCHAR(20),Salary_Amount DECIMAL(10,2),Department_Number SMALLINT,Job_Code CHAR(3))UNIQUE PRIMARY INDEX (Employee_Number)INDEX (Last_Name) ;

Primary →Secondary→

Database objects may be created ordropped as needed.

TablesViewsMacrosTriggersProcedures

CREATEDROP

Secondary indexes may be – created at table creation – created after table creation – dropped after table creation

CREATEDROP

INDEX (secondary only)



Data TypesWhen a table is created, a data type is specified for each column. Data types are dividedinto three classes -- numeric, byte, and character.The facing page shows data types.

DATE is a 32-bit integer that represents the date as yyyymmdd. It supports century andyear 2000 and is implemented with calendar-based intelligence.

TIME WITH ZONE and TIMESTAMP WITH ZONE are new ANSI standard data typesthat allow support of clock and time zone based intelligence.

DECIMAL (n,m) is a number of n digits, with m of these digits to the right of the decimalpoint.

BYTEINT is an 8-bit, signed binary whole number that may vary in range from -128 to+127.

SMALLINT is a 16-bit signed binary whole number that may vary in range from -32,768to +32,767.

INTEGER is a 32-bit signed binary whole number that may very in size from -2,147,483,648 to +2,147,483,647.

FLOAT is a 64-bit IEEE floating point number.

BYTE(n) is a fixed-length binary string of n bytes. Byte (and Varbyte) are never convertedto a different internal format. They can be used for digitized objects.

VARBYTE(n) is a variable-length binary string of n bytes.

CHAR(n) is a fixed-length character string of n characters.

VARCHAR(n) is a variable-length character string of n characters.

LONG VARCHAR is the longest variable-length character string. It is equivalent toVARCHAR(64000).

GRAPHIC, VARGRAPHIC and LONG VARGRAPHIC are the equivalent charactertypes for multi-byte character sets such as Kanji.


Data Types

TYPE Name Bytes Description

Date/Time DATE 4 YYYYMMDDTIME (WITH ZONE) 6 / 8 HHMMSSZZTIMESTAMP (WITH ZONE) 10 / 12 YYYYMMDDHHMMSSZZ

Numeric DECIMAL (18,18) 2, 4, 8 + OR -NUMERIC (18,18) 2, 4, 8 + or -BYTEINT 1 -128 to +127SMALLINT 2 -32,768 to +32,767INTEGER 4 -2,147,483,648 to +2,147,483,647FLOAT 8 IEEE floating pt

Byte BYTE(n) 0 - 64,000VARBYTE (n) 0 - 64,000

Character CHAR (n) 0 - 64,000VARCHAR (n) 0 - 64,000LONG VARCHAR same as VARCHAR (64,000)GRAPHIC 0 - 32,000VARGRAPHIC 0 - 32,000LONG VARGRAPHIC same as VARGRAPHIC(32,000)



Access Rights and PrivilegesThe diagram on the facing page shows access rights and privileges as they might be definedfor the database administrator, a programmer, a user, a system operator, and anadministrative user.

The database administrator has right to use all of the commands in the data definitionprivileges, the data manipulation privileges, and the data control privileges.

The programmer has all of those except the ability to GRANT privileges to others.

A typical user is limited to data manipulation privileges, while the operator is limited todata control privileges.

Finally, the administrative user is limited to a subset of data manipulation privileges,SELECT and EXECUTE.

Each site should carefully consider the access rules that best meet their needs.


Access Rights and Privileges

Data Definition PrivilegesCommand Object

Database and/or UserCREATE Table and/or ViewDROP Macro and/or Trigger

Stored ProcedureRole and/or Profile

Data Manipulation PrivilegesSELECTINSERT TableUPDATE ViewDELETE

EXECUTE Macro and/or Stored Procedure

Data Control PrivilegesDUMP DatabaseRESTORE TableCHECKPOINT Journal

Privileges onGRANT DatabasesREVOKE Users

Objects

A Sample Scenario

OPER

ADMIN

USER

PROGRAMMERS

DBA

Administrator

Sticky Note

No GRANT

Administrator

Sticky Note

Administrative user


Review Questions

True or False____ 1. A database will always have tables.____ 2. A user will always have a password.____ 3. A user creating a subordinate user must give up some of his/her Perm Space.____ 4. Creating tables requires the definition of at least 1 column and the user assignment of a Primary

Index.____ 5. The sum of all user and database Perm Space will equal the total space on the system.____ 6. The sum of all user and database Spool Space will equal the total space on the system.____ 7. Before a user can read a table, a table SELECT privilege must exist in the DD/D for that user.____ 8. Deleting a macro from a database reclaims Perm Space for the database.

9. Which of the following is FALSE about PERM space? ____

a. PERM space can be dynamically modified.b. The per/AMP limit of PERM space can not be exceeded.c. Maximum PERM space can be defined at the database or table level.d. Tables, index subtables, and stored procedures use PERM space.

10. Which of the following is FALSE about SPOOL space? ____

a. SPOOL space can be dynamically modified.b. The per/AMP limit of SPOOL space can not be exceeded.c. Maximum SPOOL space can be defined at the database or user level.d. Maximum SPOOL space can be defined at a value greater than the immediate parent's value.



Notes

Storing and Accessing Data Rows Page 7-1

Module 7


Storing and Accessing Data Rows


• Explain the purpose of the Primary Index

• Distinguish between Primary Index and Primary Key

• State the reasons for selecting a UPI vs. a NUPI


Page 7-2 Storing and Accessing Data Rows

Notes


Table of Contents

STORING ROWS.................................................................................................................................................... 7-4CREATING A PRIMARY INDEX ........................................................................................................................ 7-6PRIMARY INDEX VALUES ................................................................................................................................. 7-8ACCESSING VIA A UNIQUE PRIMARY INDEX ........................................................................................... 7-10ACCESSING VIA A NON-UNIQUE PRIMARY INDEX................................................................................. 7-12PRIMARY KEYS AND PRIMARY INDEXES.................................................................................................. 7-14DUPLICATE ROWS............................................................................................................................................. 7-16ROW DISTRIBUTION USING A UNIQUE PRIMARY INDEX (UPI) – CASE 1......................................... 7-18ROW DISTRIBUTION USING A NON-UNIQUE PRIMARY INDEX (NUPI) – CASE 2............................ 7-20ROW DISTRIBUTION USING A HIGHLY NON-UNIQUE PRIMARY INDEX (NUPI) – CASE 3 .......... 7-22REVIEW QUESTIONS......................................................................................................................................... 7-24



Storing RowsIdeally, the rows of every table will be distributed among all of the AMPs. There may besome circumstances where this is not true. What if there are fewer rows than AMPs?Clearly in this case, at least some AMPs will hold no rows from that table. This should beconsidered the exceptional situation, and not the rule. Each AMP is designed to hold aportion of the rows of each table. The AMP is responsible for the storage, maintenance andretrieval of the data under its control.

More ideally, the rows of each table will be evenly distributed across all of the AMPs. Thisis desirable because in operations involving all rows of the table (such as a full table scan);each AMP will have an equal portion of the work to do. When workloads are not evenlydistributed, the desired response will only be as fast as the slowest AMP.

Controlling the distribution of the rows of a table is done by the selection of the PrimaryIndex. The relative uniqueness of the Primary Index will determine the uniformity ofdistribution of the rows of this table among the AMPs.

Note that use of the word ‘AMP’ is intended to refer to both Version 1 hardware AMPs andVersion 2 AMP vprocs.


Storing Rows

Table A rowsTable B rows

AMP AMP AMP AMP

• The rows of every table are distributed among all AMPs• Each AMP is responsible for a subset of the rows of each table.• Ideally, each table will be evenly distributed among all AMPs.• Evenly distributed tables result in evenly distributed workloads.• The uniformity of distribution of the rows of a table depends on the choice of

the Primary Index.

Note:The acronym AMP is used to refer to both V1 AMPs and V2 AMP vprocs.However, this course will assume AMPs are V2 vprocs.



Creating a Primary IndexChoosing a Primary Index for a table is perhaps the most critical decision a databasedesigner makes. The choice will affect the distribution of the rows of the table and,consequently, the performance of the table in a production environment. Although manytables used combined columns as the Primary Index choice, the examples used here aresingle column indexes, mostly for the sake of simplicity.

Unique Primary Indexes (UPI’s) are desirable because they guarantee the uniformdistribution of the rows of that table.

Because it is not always feasible to pick a Unique Primary Index, it is sometimes necessaryto pick a column (or columns) which have non-unique values, that is there are duplicatevalues. This type of index is called a Non-Unique Primary Index or NUPI. While not aguarantor of uniform row distribution, the degree of uniqueness of the index will determinethe degree of uniformity of the distribution. Because all rows with the same PI value end upon the same AMP, columns with a small number of distinct values which are repeatedfrequently, typically do not make good PI candidates.

The choosing of a Primary Index is not an exact science. It requires analysis andthoughtfulness for some tables and will be completely self-evident on other tables.

The Primary Index is always designated as part of the CREATE TABLE statement. Once aPrimary Index choice has been designated for a table, it cannot be changed to somethingelse. If an alternate choice of column(s) is desired for the PI, it is necessary to drop andrecreate the table.


Creating a Primary Index

• A Primary Index is defined at table creation.• It may consist of a single column, or a combination of columns

– Limit of 16 columns with V2R4.1 and prior releases– Limit of 64 columns with V2R5.

CREATE TABLE sample_1(col_a INTEGER,col_b INTEGER,col_c INTEGER)

UNIQUE PRIMARY INDEX (col_b);

UPI If the index choice of column(s) is unique,we call this a UPI (Unique Primary Index).

A UPI choice will result in even distributionof the rows of the table across all AMPs.

CREATE TABLE sample_2(col_x INTEGER,col_y INTEGER,col_z INTEGER)

PRIMARY INDEX (col_x);

NUPI If the index choice of column(s) isn’tunique, we call this a NUPI (Non-UniquePrimary Index).

A NUPI choice will result in evendistribution of the rows of the tableproportional to the degree of uniqueness ofthe index.Note: Changing the choice of Primary Index

requires dropping and recreating the table.



Primary Index ValuesIndexes are used to access rows from a table without having to search the entire table.

On Teradata, the Primary Index is the mechanism for assigning a data row to an AMP anda location on the AMP’s disks. When a table is created it must have a Primary Indexspecified. This cannot be changed without dropping and creating the table.

Primary Indexes are very important because they have a powerful effect on the performanceof the database. The most important thing to remember is that a Primary Index is themechanism used to assign each row to an AMP and may be used to retrieve that row fromthe AMP. Thus retrievals, updates and deletes that specify the Primary Index will be muchfaster than those that don’t. Also Primary Index selection is probably the most importantfactor in the efficiency of join processing.

Earlier we learned that the Primary Key was always unique and unchanging. It was basedon the logical model of the data. The Primary Index may change and may be non-unique; itis chosen for the physical performance of the database.

There are two types of primary index – unique (UPI) and non-unique (NUPI).


Primary Index Values

• The value of the Primary Index for a specific row determines the AMPassignment for that row.

• This is done using a hashing algorithm.

PE

Row assignmentRow access

HashingAlgorithm

AMP AMP AMP

PI Value

• Accessing the row by its Primary Index value is:– always a one-AMP operation– the most efficient way to access a row

Other table access techniques:

• Secondary index access• Full table scans



Accessing Via a Unique Primary IndexA Primary Index operation is always a one-AMP operation. In the case of a UPI, the one-AMP access can return, at most, one row. In the facing example, we are looking for the rowwhose primary index value is 345. By specifying the PI value as part of our selectioncriteria, we are guaranteed that only the AMP containing the specified row will need to besearched.

The correct AMP is located by taking the PI value and passing it through a hashingalgorithm. The hashing takes place in the Parsing Engine. The output of the hashingalgorithm contains information that will point to request to a specific AMP. Once it hasisolated the appropriate AMP, finding the row is quick and efficient. How this happens wewill see in a future module.


Accessing Via a Unique Primary Index

A UPI access is a one-AMP operation which may access at most a single row.

CREATE TABLE sample_1(col_a INTEGER,col_b INTEGER,col_c INTEGER)

UNIQUE PRIMARY INDEX (col_b);

SELECT col_a,col_b,col_c

FROM sample_1WHERE col_b = 345;

PE

HashingAlgorithm

AMP

UPI = 345

AMP AMP

col_a col_b col_c

123234

col_a col_b col_c

345456

col_a col_b col_c

567678



Accessing Via a Non-Unique Primary IndexA Non-Unique Primary Index operation is also a one-AMP operation. In the case of aNUPI, the one-AMP access can return zero to many rows. In the facing example, we arelooking for the rows whose primary index value is 25. By specifying the PI value as part ofour selection criteria, we are once again guaranteeing that only the AMP containing therequired rows will need to be searched.

As before, the correct AMP is located by taking the PI value and passing it through ahashing algorithm executing in the Parsing Engine. The output of the hashing algorithm willonce again point to a specific AMP. Once it has isolated the appropriate AMP, it must nowfind all rows that have the specified value. In this example, the AMP returns two rows.


Accessing Via a Non-Unique Primary Index

A NUPI access is a one-AMP operation which may access multiple rows.

CREATE TABLE sample_2(col_x INTEGER,col_y INTEGER,col_z INTEGER)

PRIMARY INDEX (col_x);

SELECT col_x,col_y,col_z

FROM sample_2WHERE col_x = 25;

PE

HashingAlgorithm

AMP

NUPI = 25

AMP AMP

col_x col_y col_z

10 30 A 10 30 B 35 40 B

col_x col_y col_z

20 50 A 25 55 A 25 60 B

col_x col_y col_z

5 70 B 30 80 B 30 80 A

Both UPI and NUPIaccesses are oneAMP operations



Primary Keys and Primary IndexesWhile it is true that many tables use the same columns for both Primary Indexes andPrimary Keys, Indexes are conceptually different from Keys. The table on the facingpage summarizes those differences.

A Primary Key is relational data modeling term that defines, in the logical model, thecolumns that uniquely identify a row. A Primary Index is a physical databaseimplementation term that defines the actual columns used to distribute and access rows in atable.

It is also true that a significant percentage of the tables in any database will use the samecolumn(s) for both the PI and the PK. However, one should expect that in any real-worldscenario there would be some tables that will not conform to this simplistic rule. Onlythrough a careful analysis of the type of processing that will take place can the tables beproperly evaluated for PI candidates. Remember, changing your mind about the PI meansreloading the table. It may not mean never having to say you’re sorry.


Primary Keys and Primary Indexes

• Indexes are conceptually different from keys.• A PK is a relational modeling convention which allows each row to be uniquely identified.• A PI is a Teradata convention which determines how the row will be stored and accessed.• A significant percentage of tables may use the same columns for both the PK and the PI.• A well-designed database will use a PI that is different from the PK for some tables.

Primary Key Primary Index

Logical concept of data modeling Physical mechanism for access and storageTeradata doesn’t need to recognize Each table must have exactly one primary index

No limit on number of columns 16 column limit (V2R4.1); 64 column limit (V2R5)Documented in data model Defined in CREATE TABLE statement

(Optional in CREATE TABLE)Must be unique May be unique or non-unique

Identifies each row May be unique or non-uniqueValues should not change Values may be changed (Delete + Insert)

May not be NULL- requires a value May be NULLDoes not imply an access path Defines most efficient access pathChosen for logical correctness Chosen for physical performance



Duplicate RowsA duplicate row is a row of a table whose column values are all identical to another row ofthe same table. If a designer adheres to the rule that a Primary Key must be unique, then itshould preclude the possibility of having duplicate rows.

Having said that, the ANSI standard permits duplicate rows in order to satisfy therequirements of certain vendors who rely on them for certain types of auditing systems. Forexample, if I am loading a table from several different databases and the same recordappears from three different places, I might want to know that it originated from those threeplaces.

Even though this contradicts relational theory, the standard generously permits duplicaterows for these anomalous situations.

Teradata, adhering to the ANSI standard, permits duplicate rows by specifying that you wishto create a MULTISET table. In Teradata transaction mode, the default, however, is a SETtable that does not permit duplicate rows.

When MULTISET is enabled, Teradata does not do a duplicate row check for new rowsadded.

If the table is a SET table, it will only do a duplicate row check if the Primary Index is aNUPI and there are no other unique indexes on the table. If a unique index exists on thetable, duplicate index check itself will suffice to ensure there are no duplicate rows.

Also, if MULTISET is enabled, it will be overridden by choosing a UPI as the PrimaryIndex or by having a unique index (e.g., unique secondary) on another column(s) on thetable. Doing this effectively disables the MULTISET.


Duplicate Rows

A duplicate row is a row of a table whosecolumn values are all identical toanother row in the same table.

col_a col_b col_c

20 50 A 25 50 A 25 50 A

Duplicate Rows

• Because a PK uniquely identifies each row, ideally a relational table shouldnot have duplicate rows!

• The ANSI standard, however, permits duplicate rows for specializedsituations, thus Teradata permits them as well.

• You may select whether your table will or will not allow them.

* Note: If a UPI is selected on a SET table, the duplicate row check is replaced by acheck for duplicate index values.

CREATE SET TABLE table_A : :

CREATE MULTISET TABLE table_B : :

Checks for * and disallows duplicate rows. Doesn’t check for and allows duplicate rows.

The Teradata default The ANSI default



Row Distribution Using a Unique Primary Index (UPI) –Case 1

At the heart of the Teradata database is a way of predictably distributing and retrieving rowsacross AMPs. The same value stored in the same data type will always produce the samehash value. If the Primary Index is unique, Teradata can distribute the rows evenly. If thePrimary Index is slightly non-unique, that is, there are only four or five rows per indexvalue; the table will still distribute evenly. But if there are hundreds or thousands of rowsfor some index values the distribution will probably be lumpy.

In this example, the Order_Number is used as a unique primary index. Since the primaryindex value for Order_Number is unique, the distribution of rows among AMPs is veryuniform. This assures maximum efficiency because each AMP is doing approximately thesame amount of work. No AMPs sit idle waiting for another AMP to finish a task.

This way of storing the data provides for maximum efficiency and makes the best use of theparallel features of the Teradata system.


Row Distribution Using a UPI – Case 1

Notes:• Often, but not always, the PK column(s) will

be used as a UPI.• PI values for Order_Number are known to be

unique (it’s a PK).• Teradata will distribute different index

values evenly across all AMPs.• Resulting row distribution among AMPs is

very uniform.• Assures maximum efficiency for parallel

operations.

AMP AMP AMP AMP

o_# c_# o_dt o_st

7202 2 4/09 C 7415 1 4/13 C

o_# c_# o_dt o_st

7325 2 4/13 O 7103 1 4/10 O 7402 3 4/16 C

o_# c_# o_dt o_st

7188 1 4/13 C 7225 2 4/15 C

o_# c_# o_dt o_st

7324 3 4/13 O 7384 1 4/12 C

O rd erN u m b e r

C u s to m erN u m b e r

O rd e rD a te

O rd erS ta tu s

P KU P I

732573247415710372257384740271887202

231121312

4 /134 /134 /134 /104 /154 /124 /164 /134 /09

OOCOCCCCC

Order



Row Distribution Using a Non-Unique Primary Index(NUPI) – Case 2

In the example on the facing page Customer_Number has been used as a non-uniquePrimary Index (NUPI). Note row distribution among AMPs is uneven. All rows with thesame primary index value (in other words, with the same customer number) are stored onthe same AMP.

Customer_Number has three possible values, so all the rows are hashed to three AMPs,leaving the fourth AMP without rows from this table. While this distribution will work, it isnot as efficient as spreading all the rows among all the AMPs.

AMP 2 has a disproportionate number of rows and AMP 3 has none. In an all-AMPoperation AMP 2 will take longer than the other AMPs. The operation cannot complete untilAMP 2 completes its tasks. The overall operation time is increased and some of the AMPsare under-utilized.

NUPI’s can create irregular distributions, called “skewed distributions”. AMPs that havemore than an average number or rows will take longer for full table operations than the otherAMPs will. Because an operation is not complete until all AMPs have finished, this willcause the operation to finish less quickly due to being underutilized.


Row Distribution Using a NUPI – Case 2

Notes:• Customer_Number may be the preferred

access column for ORDER table, thus a goodindex candidate.

• Values for Customer_Number are somewhatnon-unique.

• Choice of Customer_Number is therefore aNUPI.

• Rows with the same PI value distribute to thesame AMP.

• Row distribution is less uniform or skewed.

o_# c_# o_dt o_st

7325 2 4/13 O 7202 2 4/09 C 7225 2 4/15 C

o_# c_# o_dt o_st

7384 1 4/12 C 7103 1 4/10 O 7415 1 4/13 C 7188 1 4/13 C

o_# c_# o_dt o_st

7402 3 4/16 C 7324 3 4/13 O

AMP AMP AMP AMP

O rd erN u m b e r


O rd e rD a te

O rd erS ta tu s

P KN U P I

732573247415710372257384740271887202

231121312

4 /134 /134 /134 /104 /154 /124 /164 /134 /09

OOCOCCCCC

Order



Row Distribution Using a Highly Non-Unique PrimaryIndex (NUPI) – Case 3

This example uses Order_Status as a NUPI. Order_Status is a poor choice, because ityields the most uneven distribution. Because there are only two possible values forOrder_Status, all of the rows are placed on two AMPs. STATUS is an example of a highly non-unique Primary Index.

When choosing a Primary Index, you should never choose a column with such a severelylimited value set. The degree of uniqueness is critical to efficiency. Choose NUPI’s thatallow all AMPs to participate fairly equally.

The degree of uniqueness of a NUPI is critical to efficiency.


Row Distribution Using a Highly Non-UniquePrimary Index (NUPI) – Case 3

O rd erN u m b e r


O rd e rD a te

O rd erS ta tu s

P KN U P I

732573247415710372257384740271887202

231121312

4 /134 /134 /134 /104 /154 /124 /164 /134 /09

OOCOCCCCC

Order Notes:• Values for Order_Status are “highly” non-

unique.• Choice of Order_Status column is a NUPI.• Only two values exist, so only two AMPs

will ever be used for this table.• Table will not perform well in parallel

operations.• Highly non-unique columns are poor PI

choices generally.• The degree of uniqueness is critical to

efficiency.

AMP AMP AMP AMP

o_# c_# o_dt o_st

7402 3 4/16 C 7202 2 4/09 C 7225 2 4/15 C 7415 1 4/13 C 7188 1 4/13 C 7384 1 4/12 C

o_# c_# o_dt o_st

7103 1 4/10 O 7324 3 4/13 O 7325 2 4/13 O


Review Questions

For each statement, indicate whether it applies to:

UPI’s, NUPI’s, or Either

_______ 1. Specified in CREATE TABLE statement_______ 2. Provides uniform distribution via the hashing algorithm_______ 3. May be up to 64 columns in V2R5_______ 4. Always a one-AMP operation_______ 5. Access will return (at most) a single row_______ 6. Used to assign a row to a specific AMP_______ 7. Allows a null or nulls_______ 8. Required on every table_______ 9. Permits duplicate rows_______ 10. Used as a Primary Key implementation



Notes

Primary Index Mechanics Page 8-1

Module 8


Primary Index Mechanics


• Explain the role of the hashing algorithm and the hash map inlocating a row.

• Explain the makeup of the Row ID and its role in row storage.

• Describe the sequence of events for locating a row given its PIvalue.


Page 8-2 Primary Index Mechanics

Notes


Table of Contents

HASHING PRIMARY INDEX VALUES ............................................................................................................. 8-4HASHING DOWN TO THE AMPS ...................................................................................................................... 8-6A HASHING EXAMPLE........................................................................................................................................ 8-8THE HASH MAP................................................................................................................................................... 8-10IDENTIFYING ROWS ......................................................................................................................................... 8-12THE ROW ID ........................................................................................................................................................ 8-14STORING ROWS (1 OF 2)................................................................................................................................... 8-16

STORING ROWS (2 OF 2)....................................................................................................................................... 8-18LOCATING A ROW ON AN AMP USING A PI............................................................................................... 8-20REVIEW QUESTIONS......................................................................................................................................... 8-22



Hashing Primary Index ValuesA hashing algorithm is a standard data processing technique that takes in a data value, likelast name or order number, and systematically mixes it up so that the incoming values areconverted to a number in a range from zero to the specified maximum value. A successfulhashing scheme scatters the input evenly over the range of possible output values.

It is predictable in that Smith will always hash to the same value and Jones will always hashto another (hopefully different) value. With a good hashing algorithm any patterns in theinput data should disappear in the output data. If many names begin with “S”, they shouldand will not all hash to the same group of hash values. If order numbers all have “00” in thehundreds and tens place or if all names are four letters long we should still see the hashvalues spread fairly evenly over the whole range.

Textbooks still say that this requires manually designing and tuning a hash algorithm foreach new type of data values. However, the Teradata algorithm works predictably well overany data, typically loading each AMP with variations in the range of .1% to .5% betweenAMPs. For extremely large systems, the variation can be as low as .001% between AMPs.

Teradata also uses hashing quite differently than other data storage systems. Other hasheddata storage systems equate a bucket with a physical location on disk. In Teradata, a bucketis simply an entry in a hash map. Each hash map entry points to a single AMP. Therefore,changing the number of AMPs does not require any adjustment to the hashing algorithm.Teradata simply adjusts the hash maps and redistributes any affected rows.

The hash maps must always be available to the Message Passing Layer. In Version 1, a hashmap had 3643 buckets (entries), while in Version 2 a hash map has 65,536 entries.

When the hash bucket has determined the destination AMP, the full 32-bit row hash plus theTable-ID is used to assign the row to a cylinder and a data block on the AMPs disk storage.In Version 1, the hashing algorithm can produce over 238,000,000 row hash values, while inVersion 2 it can produce over 4,000,000,000 row hash values.


Hashing Primary Index Values

Hashing Algorithm

RH Data

Row Hash PI valuesDSW and data

PARSER

Data Table

Message Passing Layer (Hash Maps)

AMP 1 AMP n - 1AMP x... ...AMP 0 AMP n

PI value = 38

Hashing Algorithm

1177 7C3C

SQL with primary index valuesand data.

For example: Assume PI value is 38

Summary

The MPL uses the DSW of1177 and uses this value tolocate bucket #1177 in theHash Map.

Bucket# 1177 contains theAMP number that has thishash value - effectively theAMP with this row.

DSWHash Maps

AMP #

Row ID Row Data Row Hash Uniq Value

x '00000000'

x'1177 7C3C' 0000 0001 38

x 'FFFFFFFF'



Hashing Down to the AMPsThe rows of all tables are distributed across the AMPs according to their Primary Indexvalue. The Primary Index value goes into the hashing algorithm and the output is a 32-bitRow Hash. The high order 16 bits are referred to as the “bucket number” and are used toidentify a hash map entry. The “hash bucket” is also referred to as then DSW – DestinationSelection Word. This entry, in turn, is used to identify the AMP that will be targeted. Theremaining 16 bits are not used to locate the AMP.

The entire 32-bit Row Hash is used by the selected AMP to locate the row within its diskspace.

Hash maps are uniquely configured for each size of system, thus a 96 AMP system willhave a hash map different from a 64 AMP system, but another 64 AMP system will have thesame map.

Each hash map is simply an array that associates DSW values (or bucket numbers) withspecific AMPs.

When a system grows, new AMPs are typically added. This requires a change to the hashmap to reflect the new total number of possible target AMPs.


Hashing Down to the AMPsIndex value(s)

hashing algorithm

Hash Map

AMP #

The hashing algorithm is designed to insure even distribution ofunique values across all AMPs.

Different hashing algorithms are used for different internationalcharacter sets.

A Row Hash is the 32-bit result of applying a hashing algorithm toan index value.

The DSW or Hash Bucket is represented by the high order 16 bitsof the Row Hash.

A Hash Map is uniquely configured for each system.

It is a array of 65,536 entries (buckets) which associates bucketnumbers with specific AMPs.

Two systems with the same number of AMPs will have the sameHash Map.

Changing the number of AMPs in a system requires a change tothe Hash Map.

{{{{

DSW orHash Bucket #

Row Hash



A Hashing ExampleThe facing page shows an example of how the hashing algorithm would produce a 32-bitrow hash value on the primary index value “7202”.

The hash value is divided into two parts. The first 16 bits are the Destination SelectionWord (DSW). In Version 2, the full 16 bits of the DSW is used for the Hash Bucket. InVersion 1, the DSW is further divided into the Hash Map number (4 bits) and the HashBucket number (12 bits). The last 16 bits of the hash value are the Remainder. In Version1, Hash map number distinguishes normal row from fallback copy and normal distributionfrom reconfiguration.

In Version 2, this is accomplished elsewhere. The hash bucket points to a particular hashmap entry, which in turn points to one AMP. The entire Row Hash along with the Table IDreferences a particular logical location on that AMP.


A Hashing Example

OrderOrder

NumberPKUPI

CustomerNumber

OrderDate

OrderStatus

7325 2 4/13 O7324 3 4/13 O7415 3 4/13 O7415 1 4/13 C7103 1 4/10 O7225 2 4/15 C7384 1 4/12 C7402 3 4/12 C7188 1 4/13 C7202 2 4/09 C

SELECT * FROM orderWHERE order_number = 7202;

7202

Hashing Algorithm

691B 14AE

32 bit Row Hash Remaining 16 bitsDestination Selection Word

0110 1001 0001 1011 0001 0100 1010 1110

6 9 1 B



The Hash MapA hash map is simply an array of 65,536 entries where each entry is two bytes long. Thehash map is loaded into memory and is used by Teradata software. Each entry contains anAMP number for the system on which Teradata is implemented. The DSW (DestinationSelection Word) or bucket number is an offset into the hash map to locate a specific entry(or AMP).

To determine the destination AMP for a Primary Index operation, the hash map is checkedby BYNET software using the row hash information. A message is placed on the BYNETto be sent to the target AMP using point-to-point communication.

In the example, the DSW entry 691B (hexadecimal) contains an entry that identified AMP 7.AMP 7 will be the recipient of the message from the Message Passing Layer.

One major difference between Version 1 and Version 2 is the increase in the number of hashbuckets from 3643 to 65,536. This results in fewer hash collisions, better performance, andthe ability to configure systems with thousands of AMPs rather than hundreds of AMPs.


The Hash Map

7202 Hashing Algorithm

(Hexadecimal)

691B 14AE

HASH MAP

00 01 02 03 04 05 06 07 08 09 00 01 02 03 04 0506 07 08 09 00 01 02 03 04 05 06 07 08 09 00 0102 03 04 05 06 07 08 09 00 01 02 03 04 05 06 0708 09 00 01 02 03 04 05 06 07 08 09 00 01 02 0304 05 06 07 08 09 00 01 02 03 04 05 06 07 08 0900 01 02 03 04 05 06 07 08 09 00 01 02 03 04 05

0 1 2 3 4 5 6 7 8 9 A B C D E F690691692693694695

32 bit Row Hash Remaining 16 bitsDestination Selection Word

0110 1001 0001 1011 0001 0100 1010 1110

6 9 1 B

AMP 7

7202 2 4/09 C



Identifying RowsCan two different PI values come out of the hashing algorithm with the same row hashvalue? The answer is “Yes”. There are two ways that can happen.

First, two different primary index values may happen to hash identically. This is called ahash synonym.

Secondly, if a non-unique primary index is used, duplicate NUPI values will produce thesame row hash.


Identifying Rows

Consideration #1

A Row Hash = 32 bits = 4.2 billion possiblevalues

Because there is an infinite number ofpossible data values, some data values willhave to share the same row hash.

Hash Algorithm

1254 7769

10A2 2936 10A2 2936 Hash Synonyms

Data values input

Consideration #2

A Primary Index may be non-unique (NUPI).

Different rows will have the same PI valueand thus the same row hash.

A row hash is not adequate to uniquely identify a row.

ConclusionA row hash is not adequate to uniquely identify a row.

Hash Algorithm

(John)'Smith'

0016 5557

(Dave)'Smith' NUPI Duplicates

Rows havesame hash

0016 5557



The Row IDIn order to differentiate each row in a table, every row is assigned a unique Row ID. TheRow ID is a combination of the row hash value plus a uniqueness value. The AMPappends the uniqueness value to the row hash when it is inserted. The Uniqueness Value isused to differentiate between PI values that generate identical row hashes.

The first row inserted with a particular row hash value is assigned a uniqueness value of 1.Each new row with the same row hash is assigned an integer value one greater than thecurrent largest uniqueness value for this Row ID.

If a row is deleted or the primary index is modified, the uniqueness value can be reused.

Only the Row Hash portion is used in Primary Index operations. The entire Row ID is usedfor Secondary Index support that is discussed in a later module.


The Row ID

To uniquely identify a row, we add a 32-bit uniqueness value.The combined row hash and uniqueness value is called a Row ID.

Row Hash(32 bits)

Uniqueness Id(32 bits)

Row ID

Each stored rowhas a Row ID as aprefix.

Rows are logicallymaintained in RowID sequence.

Row ID Row Data

3B11 5032 0000 0001 1018 Reynolds Jane3B11 5032 0000 0002 1020 Davidson Evan3B11 5032 0000 0003 1031 Green Jason3B11 5033 0000 0001 1014 Jacobs Paul3B11 5034 0000 0001 1012 Chevas Jose3B11 5034 0000 0002 1021 Carnet Jean : : : : :

Row Hash Unique ID Emp_No Last_Name First_Name

Row ID Row Data



Storing Rows (1 of 2)Rows are stored in a data block in Row ID sequence. As rows are added to a table with thesame row hash, the uniqueness value is incremented by one in order to provide a uniqueRow ID..

Assume Last_Name is a NUPI and that all rows in this example hash to the same AMP.

The ‘John Smith’ row is assigned to AMP 3 based on the bucket number portion of the rowhash. Because it is the first row with this row hash, a uniqueness id of 1 is assigned.

The ‘Sam Adams’ row has a different row hash and thus is also assigned a uniqueness valueof 1. The bucket number, although different, also points to AMP 3 in the hash map.


Storing Rows (1 of 2)

Assumptions:Last_Name is defined as a NUPI.All rows in this example hash to the same AMP.

Add a row for 'John Smith'

Add a row for 'Sam Adams'

Row ID Row Data Row Hash Unique ID Last_Name First_Name Etc.

0016 5557 0000 0001 Smith John

Row ID Row Data Row Hash Unique ID Last_Name First_Name Etc.0016 5557 0000 0001 Smith John1058 9829 0000 0001 Adams Sam

'Adams' Hash Algorithm 1058 9829 Hash Map AMP #3

'Smith' Hash Algorithm 0016 5557 Hash Map AMP #3



Storing Rows (2 of 2)The ‘Fred Smith’ row hashes to the same row hash as ‘John Smith’ because it is a NUPIduplicate. It is therefore assigned a uniqueness id of 2.

The ‘Dan Jones’ row also hashes to the same row hash because it is a hash synonym. It isthus assigned a uniqueness id of 3.


Storing Rows (2 of 2)

Add a row for 'Fred Smith' - (NUPI Duplicate)

Row ID Row Data Row Hash Unique ID Last_Name First_Name Etc.0016 5557 0000 0001 Smith John0016 5557 0000 0002 Smith Fred1058 9829 0000 0001 Adams Sam

'Smith' Hash Algorithm 0016 5557 Hash Map AMP #3

Add a row for 'Dan Jones' - (Hash Synonym)

'Jones' Hash Algorithm 0016 5557 Hash Map AMP #3

Row ID Row Data Row Hash Unique ID Last_Name First_Name Etc.0016 5557 0000 0001 Smith John0016 5557 0000 0002 Smith Fred0016 5557 0000 0003 Jones Dan1058 9829 0000 0001 Adams Sam

Given the row hash, what other information would be needed to find the 'Dan Jones' row? … The 'Fred Smith' row?



Locating a Row on an AMP Using a PITo locate a row, the AMP file system searches through a memory-resident structure calledthe Master Index. An entry in the Master Index will indicate that if a row with this Table IDand row hash exists, then it must be on a specific disk cylinder.

The file system will then search through the designated Cylinder Index. There it will find anentry that indicates that if a row with this Table ID and row hash exists, it must be in onespecific data block on that cylinder.

The file system then searches the data block until it locates the row(s) or returns a No RowsFound condition code.

Table-idRow-hash

Row-hashPI Value

MasterIndex

CylinderIndex

DataBlock Data Row

Table-idRow-hash


Locating a Row On An AMP Using a PI

Locating a row on an AMPrequires three input elements:

1. The Table ID2. The Row Hash of the PI3. The PI value itself

Cyl 1Index

Cyl 2Index

Cyl 3Index

Cyl 4Index

Cyl 5Index

Cyl 6Index

Cyl 7Index

Master

Index

Data RowData Row

DATA BLOCK

AMP #3

Cylinder #PI Value

MasterIndex

CylinderIndex

DataBlock

Table IdRow Hash

Table IdRow HashCylinder #

Row HashPI Value

Cylinder #

Data Block Address

Data Row

START WITH: FIND:APPLY TO:

Table IDRow Hash

Administrator

Sticky Note

It should be Uniqueness Value


Review Questions

Fill in the Blanks

1. The output of the hashing algorithm is called the _____ _____.

2. To determine the target AMP, the Message Passing Layer must lookup an entry in theHash Map based on the ________ number.

3. Two different PI values which hash to the same values are called Hash ___________ .

4. A Row ID consists of a row hash plus a ____________ value.

5. A uniqueness value is required to produce a unique Row ID because of ________________ and ______ ___________ .

6. Once the target AMP has been determined for a PI search, the _______ ________ for thatAMP must be consulted.

7. The Cylinder Index points us to the address and length of the data _______ .



Notes

Secondary Indexes and Table Scans Page 9-1

Module 9


Secondary Indexes and Table Scans


• Define Secondary Indexes.

• Distinguish between the implementation of unique andnon-unique secondary indexes.

• Define Full Table Scans and what causes them.

• Describe the operation of a Full Table Scan in a parallelenvironment.


Page 9-2 Secondary Indexes and Table Scans

Notes


Table of Contents

SECONDARY INDEXES ....................................................................................................................................... 9-4CHOOSING A SECONDARY INDEX.................................................................................................................. 9-6UNIQUE SECONDARY INDEX (USI) ACCESS................................................................................................. 9-8NON-UNIQUE SECONDARY INDEX (NUSI) ACCESS.................................................................................. 9-10COMPARISON OF PRIMARY AND SECONDARY INDEXES..................................................................... 9-12FULL TABLE SCANS .......................................................................................................................................... 9-14REVIEW QUESTIONS......................................................................................................................................... 9-16



Secondary IndexesA secondary index is an alternate path to the data. Secondary Indexes are used to improveperformance by allowing the user to avoid scanning the entire table. A Secondary Index islike a Primary Index in that it allows the user to locate rows. It is unlike a Primary Index inthat it has no influence on the way rows are distributed among AMPs. A database designertypically chooses a secondary index because it provides faster set selection.

Primary Indexes requests require the services on only one AMP to access rows, whilesecondary indexes require at least two and possibly all AMPs, depending on the index andthe type of operation. A secondary index search will typically be less expensive than a fulltable scan.

Secondary indexes add overhead to the table, both in terms of disk space and maintenance,however they may be dropped when not needed, and recreated whenever they would behelpful.


Secondary Indexes

• A secondary Index provides an alternate path to the rows of a table.

• A table can have from 0 to 32 secondary indexes.

• Secondary Indexes:

– Do not effect table distribution.

– Add overhead, both in terms of disk space and maintenance.

– May be added or dropped dynamically as needed.

– Are chosen to improve table performance.

There are 3 general ways to access a table:

Primary Index access (one AMP access)

Secondary Index access (two or all AMP access)

Full Table Scan (all AMP access)



Choosing a Secondary IndexJust as with primary indexes, there are two types of secondary indexes – unique (USI) andnon-unique (NUSI).

Secondary Indexes may be specified at table creation or at any time during the life of thetable. It may consist of up to 16 columns, however to get the benefit of the index, the querywould have to specify a value for all 16 values.

Unique Secondary Indexes (USI) have two possible purposes. They can speed up accessto a row which otherwise might require a full table scan without having to rely on theprimary index. Additionally, they can be used to enforce uniqueness on a column or set ofcolumns. This is sometimes the case with a Primary Key which is not designated as thePrimary Index. Making it a USI has the effect of enforcing the uniqueness of the PK.

Non-Unique Secondary Indexes (NUSI) are usually specified in order to prevent full tablescans. NUSI’s however do activate all AMPs - after all, the value being sought might welllive on many different AMPs (only Primary Indexes have same values on same AMPs). Ifthe optimizer decides that the cost of using the secondary index is greater than a table scanwould be, it opts for the table scan.

All secondary indexes cause an AMP local subtable to be built and maintained as columnvalues change. Secondary index subtables consist of rows which associate the secondaryindex value with one or more rows in the base table. When the index is dropped, thesubtable is physically removed.


Choosing a Secondary Index

A Secondary Index may be defined ...– at table creation (CREATE TABLE)– following table creation (CREATE INDEX)– may be up to 16 columns (V2R4.1); 64 columns (V2R5.0)

If the index choice of column(s) isunique, it is called a USI.

Unique Secondary Index)

Accessing a row via a USI is a 2 AMPoperation.

USIIf the index choice of column(s) is non-unique, it is called a NUSI.

Non-Unique Secondary Index

Accessing row(s) via a NUSI is an allAMP operation.

NUSI

CREATE UNIQUE INDEX (Employee_Number) ON Employee;

CREATE INDEX (Last_Name) ON Employee;

Notes:• Secondary Indexes cause an internal sub-table to be built.• Dropping the index causes the sub-table to be deleted.



Unique Secondary Index (USI) AccessThe facing page shows the two AMP accesses necessary to retrieve a row via a UniqueSecondary Index access.

After the row hash of the secondary index value is calculated, the hash map points us toAMP 2 as containing the subtable row for this USI value. After locating the subtable row inAMP 2, we find the row-id of the base row we are seeking. This base row id (whichincludes the row hash) again allows the hash map to point us to AMP 4 which contains thebase row.

Secondary index access uses the complete row-id to locate the row, unlike primary indexaccess, which only uses the row hash portion.

The Customer table below is the table used in the example. It is only a partial listing of therows.

Customer Table

Cust Name Phone

USI NUPI

37987495275645

WhiteBrownSmithPetersJonesSmithAdams

555-4444333-9999555-6666555-7777222-8888555-7777444-6666


Unique Secondary Index (USI) Access

CREATE UNIQUE INDEX(Cust) ON Customer;

SELECT *FROM CustomerWHERE Cust = 56;

Create USI

Access via USI

HashingAlgorithm

USI Value = 56

PETable ID

100

Row Hash

778

Unique Val

7

AMP 1 AMP 2 AMP 3 AMP 4

Base Table Base Table Base Table Base Table

RowID Cust Name PhoneUSI NUPI

471, 1 45 Adams 444-6666555, 6 98 Brown 333-9999717, 2 72 Adams 666-7777884, 1 74 Smith 555-6666


147, 1 49 Smith 111-6666147, 2 12 Young 777-4444388, 1 27 Jones 222-8888822, 1 62 Black 444-5555


107, 1 37 White 555-4444536, 5 84 Rice 666-5555638, 1 31 Adams 111-2222640, 1 40 Smith 222-3333


639, 1 77 Jones 777-6666778, 3 95 Peters 555-7777778, 7 56 Smith 555-7777915, 9 51 Marsh 888-2222

USI Subtable USI Subtable USI Subtable USI Subtable

RowID Cust RowID244, 1 74 884, 1505, 1 77 639, 1744, 4 51 915, 9757, 1 27 388, 1

RowID Cust RowID135, 1 98 555, 6296, 1 84 536, 5602, 1 56 778, 7969, 1 49 147, 1

RowID Cust RowID288, 1 31 638, 1339, 1 40 640, 1372, 2 45 471, 1588, 1 95 778, 3

RowID Cust RowID175, 1 37 107, 1489, 1 72 717, 2838, 4 12 147, 2919, 1 62 822, 1



Message Passing LayerCustomerTable ID = 100

Table ID Row Hash USI Value

100 602 56

to MPL



Non-Unique Secondary Index (NUSI) AccessThe facing page shows an all-AMP access necessary to retrieve a row via a Non-UniqueSecondary Index access.

After the row hash of the secondary index value is calculated, the Message Passing Layerwill automatically activate all AMPs per instructions of the Parsing Engine. Each AMPlocates the subtable rows containing the qualifying value and row hash. These subtablerows contain the row-id(s) for the base rows, which are guaranteed to be on the same AMPas the subtable row. This reduces activity in the MPL and essentially makes the query anAMP-local operation.

Because each AMP may have more than one qualifying row, it is possible for the subtablerow to have more than row-id for the base table rows.

The Customer table below is the table used in the example. It is only a partial listing of therows.

Customer Table

Cust Name Phone

NUSI NUPI

37987495275645

WhiteBrownSmithPetersJonesSmithAdams

555-4444333-9999555-6666555-7777222-8888555-7777444-6666


Non-Unique Secondary Index (NUSI) Access

CREATE INDEX (Name) ONCustomer;

SELECT *FROM CustomerWHERE Name = 'Adams';

Create NUSI

Access via NUSI

HashingAlgorithm

NUSI Value = 'Adams'

PE



CustomerTable ID = 100

Table ID Row Hash NUSI Value

100 567 Adams

to MPL

NUSI Subtable NUSI Subtable NUSI Subtable NUSI Subtable

RowID Name RowID432, 8 Smith 640, 1448, 1 White 107, 1567, 3 Adams 638, 1656, 1 Rice 536, 5

RowID Name RowID432, 1 Smith 147, 1448, 4 Black 822, 1567, 6 Jones 338, 1770, 1 Young 147, 2

RowID Name RowID155, 1 Marsh 915, 9396, 1 Peters 778, 3432, 5 Smith 778, 7567, 1 Jones 639, 1

RowID Name RowID432, 3 Smith 884, 1567, 2 Adams 471, 1

717, 2852, 1 Brown 555, 6


Base Table Base Table Base Table Base Table

RowID Cust Name Phone NUSI NUPI

471, 1 45 Adams 444-6666555, 6 98 Brown 333-9999717, 2 72 Adams 666-7777884, 1 74 Smith 555-6666


147, 1 49 Smith 111-6666147, 2 12 Young 777-4444388, 1 27 Jones 222-8888822, 1 62 Black 444-5555


107, 1 37 White 555-4444536, 5 84 Rice 666-5555638, 1 31 Adams 111-2222640, 1 40 Smith 222-3333


639, 1 77 Jones 777-6666778, 3 95 Peters 555-7777778, 7 56 Smith 555-7777915, 9 51 Marsh 888-2222



Comparison of Primary and Secondary IndexesThe table on the facing page compares and contrasts primary and secondary indexes:

Primary indexes are required; secondary indexes are optional. All tables must have amethod of distributing rows among AMPs -- the Primary Index.

A table can only have one primary index, but it can have up to 32 secondary indexes.

Both primary and secondary indexes can have up to 16 columns.

Secondary indexes, like primary indexes, can be either unique (USI) or non-unique (NUSI).

The secondary index does not affect the distribution of rows. Rows are only distributedaccording to the Primary Index values.

Secondary indexes can be created and dropped dynamically. In other words, SecondaryIndexes can be added as needed. In fact, in some cases it is a good idea to wait and see howthe database is used and then add Secondary Indexes to facilitate that usage.

Both primary and secondary indexes affect system performance. However, Primary andSecondary Indexes affect performance for different reasons. A poorly-chosen PI results in“lumpy” data distribution which makes some AMPs do more work than others and slows thesystem.

Secondary Indexes affect performance because they require subtables. Both indexes allowrapid retrieval of specific rows.

Both primary and secondary indexes can be created using multiple data types.

Secondary indexes are stored in separate subtables; primary indexes are not.

Because secondary indexes require separate subtables, extra I/O is needed to maintain thosesubtables.


Comparison of Primary and Secondary Indexes

Index Feature Primary Secondary

Required? Yes NoNumber per Table 1 0 - 32Max Number of Columns (V2R4.1) 16 16Max Number of Columns (V2R5) 64 64Unique or Non-unique Both BothAffects Row Distribution Yes NoCreated/Dropped Dynamically No YesImproves Access Yes YesMultiple Data Types Yes YesSeparate Physical Structure No Sub-tableExtra Processing Overhead No YesMay be Partitioned (V2R5) Yes No



Full Table ScansA full table scan is another way to access data without using any Primary or SecondaryIndexes.

In evaluating an SQL request, the Parser examines all possible access methods and choosesthe one it believes to be the most efficient. The coding of the SQL request along with thedemographics of the table and the availability of indexes all play a role in the decision of theParser. Some coding constructs, listed on the facing page, always cause a full table scan. Inother cases, it might be chosen because it is the most efficient method. In general, if thenumber of physical reads exceeds the number of data blocks then the optimizer may decidethat a full-table scan is faster.

With a full table scan, each data block is found using the Master and Cylinder Indexes andeach data row is accessed only once.

As long as the choice of Primary Index has caused the table rows to distribute evenly acrossall of the AMPs, the parallel processing of the AMPs can do the full table scan quickly. Thefile system keeps each table on as few cylinders as practical to help reduce the cost full tablescans.

While full table scans are impractical and even disallowed on some systems, the Teradataroutinely permits ad hoc queries with full table scans.


Full Table Scans

Every row of the table must be read.All AMPs scan their portion of the table in parallel.Fast and efficient on Teradata due to parallelism.Full table scans typically occur when either:

– An index is not used in the query– An index is used in a non-equality test

Cust_ID Cust_Name Cust_PhoneUSI NUPI

Customer

SELECT * FROM Customer WHERE Cust_Phone LIKE '524-_ _ _ _';

SELECT * FROM Customer WHERE Cust_Name = 'Davis';

SELECT * FROM Customer WHERE Cust_ID > 1000;

Examples of Full Table Scans:


Review Questions

USIAccess FTS

# AMPs

# rows

Parallel Operation

Uses Hash Maps

Uses Separate Sub-table

Reads all data blocks of table

Fill each box with either Yes, No, or the appropriatenumber.

NUSIAccess



Notes

Data Protection Page 10-1

Module 10


Data Protection


• Explain the concept of FALLBACK tables.

• List the types and levels of locking provided by Teradata.

• Describe the Recovery, Transient and Permanent Journalsand their function.

• List the utilities available for archive and recovery.


Page 10-2 Data Protection

Notes


Table of Contents

DATA PROTECTION FEATURES .................................................................................................................... 10-4DISK ARRAYS...................................................................................................................................................... 10-6RAID TECHNOLOGIES...................................................................................................................................... 10-8RAID 1 – MIRRORING...................................................................................................................................... 10-10

RAID 1 – STRIPED MIRRORING......................................................................................................................... 10-10RAID 1 SUMMARY............................................................................................................................................ 10-12RAID 5 – DATA PARITY PROTECTION....................................................................................................... 10-14RAID 5 SUMMARY............................................................................................................................................ 10-16TERADATA – RAID 1 AND RAID 5 ................................................................................................................ 10-18CLIQUES ............................................................................................................................................................. 10-20TERADATA VPROC MIGRATION................................................................................................................. 10-22LOCKS ................................................................................................................................................................. 10-24LOCKING MODIFIER ...................................................................................................................................... 10-26

ACCESS ........................................................................................................................................................... 10-26NOWAIT .......................................................................................................................................................... 10-26

RULES OF LOCKING ....................................................................................................................................... 10-28ACCESS LOCKS................................................................................................................................................. 10-30FALLBACK ......................................................................................................................................................... 10-32FALLBACK CLUSTERS ................................................................................................................................... 10-34FALLBACK AND RAID PROTECTION......................................................................................................... 10-36FALLBACK AND RAID 1 EXAMPLE............................................................................................................. 10-38

FALLBACK AND RAID 1 EXAMPLE (CONT.)....................................................................................................... 10-40FALLBACK AND RAID 1 EXAMPLE (CONT.)....................................................................................................... 10-42FALLBACK AND RAID 1 EXAMPLE (CONT.)....................................................................................................... 10-44FALLBACK AND RAID 1 EXAMPLE (CONT.)....................................................................................................... 10-46

FALLBACK VS. NON-FALLBACK TABLES SUMMARY .......................................................................... 10-48CLUSTERS AND CLIQUES.............................................................................................................................. 10-50RECOVERY JOURNAL FOR DOWN AMPS................................................................................................. 10-52TRANSIENT JOURNAL.................................................................................................................................... 10-54PERMANENT JOURNAL ................................................................................................................................. 10-56ARCHIVING AND RECOVERING DATA ..................................................................................................... 10-58REVIEW QUESTIONS....................................................................................................................................... 10-60



Data Protection FeaturesDisk Arrays – Disk arrays provide RAID 1, RAID 5, or RAID S data protection. If a diskdrive fails, the array subsystem provides continuous access to the data. Systems with diskarrays are configured with redundant SCSI adapters, SCSI buses, and array controllers toprovide highly available access to the data.

Clique - a set of Teradata nodes that share a common set of disk arrays. In the event ofnode failure, all vprocs can migrate to another available node in the clique. All nodes in theclique must have access to the same disk arrays.

Locks - Locking prevents multiple users who are trying to change the same data at the sametime from violating the data's integrity. This concurrency control is implemented by lockingthe desired data. Locks are automatically acquired during the processing of a request andreleased at the termination of the request. In addition, users can specify locks. There arefour types of locks: Exclusive, Write, Read, and Access.

Fallback - protects your data by storing a second copy of each row of a table on analternative “fallback AMP”. If an AMP fails, the system accesses the fallback rows to meetrequests. Fallback provides AMP fault tolerance at the table level. With Fallback tables, ifone AMP fails, all of the table data is still available. Users may continue to use Fallbacktables without any loss of available data.

Down-AMP Recovery Journal - started automatically when the system has a failed ordown AMP. Its purpose is to log any changes to rows which reside on the down AMP.

Transient Journal - exists to permit the successful rollback of a failed transaction.Transactions are not committed to the database until an End Transaction request has beenreceived by the AMPs, either implicitly or explicitly. Until that time, there is always thepossibility that the transaction may fail in which case the participating table(s) must berestored to their pre-transaction state.

Permanent Journal - provides selective or full database recovery to a specified point intime by keeping either before-image or after-images of rows in a journal. It permitsrecovery from unexpected hardware or software disasters.

ARC and NetBackup/NetVault - ARC command scripts provide the capability to backupand restore the Teradata database. The NetBackup and NetVault utilities provide a GUIbased front-end for creation and execution of ARC command scripts.


Data Protection Features

Facilities that provide system-level protection

Disk Arrays– RAID data protection (e.g., RAID 1)– Redundant SCSI buses and array controllers

Cliques and Vproc Migration– SMP or O.S. failures - Vprocs can migrate to other nodes within the clique.

Facilities that provide Teradata DB protectionLocks – provides data integrityFallback – provides data access with a “down” AMPDown AMP Recovery Journal – fast recovery of fallback rows for AMPsTransient Journal – automatic rollback of aborted transactionsPermanent Journal – optional before and after-image journalingARC – Archive/Restore facilityNetBackup and NetVault – provide tape management and ARC script

creation and scheduling capabilities



Disk ArraysDisk arrays utilize a technology called RAID. RAID or Redundant Array of IndependentDisks is an storage technology with the potential to revolutionize the way on-line data isstored in computers. Spanning the entire spectrum from personal computers to mainframes,disk arrays (utilizing RAID technology) offer significant improvements in availability,reliability and maintainability of information storage, along with higher performance thantoday’s conventional disks. Yet the concept behind disk arrays is relatively simple.

A disk array subsystem consists of controller(s) which drive a set of disks (multiples of 5with NCR 6250 disk arrays, multiples of 4 with NCR 6285 disk arrays). Typically, a diskarray is configured to represent a number of logical volumes (or disks), each of whichappears to be a physical disk to the user. A logical volume can be configured to reside onmultiple physical disks. The fact that a logical volume is located on 1 or more disks istransparent to the user.

There is one immediate advantage of having the data spread across a number of individualseparate disks which arises from the redundant manner in which the data can be stored in thedisk array. The remarkable benefit of this feature is that if any single disk in the array fails,the unit continues to function without loss of data. This is possible because redundancyinformation is stored separate from the data. The redundancy information, as will beexplained, can be a copy of the data or other information that can be used to reconstruct anydata that was stored on a failed disk.

Secondly, performance increases for specific applications are possible as the effective seektime for finding records on a given disk can potentially be reduced by allowing multiplesimultaneous accesses of different blocks on different disks. Alternatively, with a differentarchitecture, the rate at which data is transferred to and from the disk array can be increasedsignificantly over that of a single disk utilizing parallel reads and writes of the data spreadacross the disks in the array. This function is referred to as “striping the data”.

Finally, disk array subsystem maintenance is typically simplified because it is possible toreplace (“hot swap”) individual disks and other components while the system continues tofunction. You no longer have to bring down the system to replace a disk.


Disk Arrays

DAC

DAC

Host Operating System

Utilities Applications

Why Disk Arrays?• High availability through data mirroring or data parity protection.• Better I/O performance through implementation of RAID technology at the

hardware level.• Convenience - automatic disk recovery and data reconstruction when

mirroring or data parity protection is used.



RAID TechnologiesRAID is an acronym for Redundant Array of Independent Disks. The term was coined in1988 in a paper describing array configuration and application by researchers and authorsPatterson, Gibson and Katz of the University of California at Berkeley. The word redundantimplies that either data, functions and/or components have been duplicated in the array’sarchitecture. Duplication of data, functions, and hardware ensures that even in the event of afailed drive or other components, data is not lost and is continuously available.

The industry currently has agreed upon six RAID configuration levels and designated themas RAID 0 through RAID 5. The physical configuration is dictated to some extent by thechoice of RAID level; however, RAID conventions specify more precisely how data isstored on disk.

RAID 0 Data stripingRAID 1 Disk mirroringRAID 2 Parallel array, hamming codeRAID 3 Parallel array with parityRAID 4 Data parity protection, dedicated parity driveRAID 5 Data parity protection, interleaved parity

With Teradata, the two RAID technologies most commonly used are RAID 1, which ismirroring, and RAID 5 (or RAID S), which is data parity protection. A brief description ofRAID 5 and RAID S will be provided in this module.

There are other RAID technologies that are defined by specific vendors or are accepted inthe data processing industry.

For example, RAID 10 or RAID 1+0 (or RAID 0+1) is considered to be “striped mirroring”.

RAID S is a RAID technology used with older EMC2 disk arrays. RAID S provides dataparity protection similar to a RAID 5 environment. The main difference is that parity iscalculated at the drive level rather than at the array controller.

The Mylex Disk Array Adapter is a PCI adapter board that provides RAID 0, RAID 1,RAID 5, RAID 6, and RAID 7 capabilities with a set of internal disks. This board iscommonly used with the NCR 4300 and 4400 servers.

RAID level classifications do not imply superiority of one mode over another. Each modehas its rightful application. In fact, these modes of operation can be combined within asingle system configuration, within product limitations, to obtain maximum flexibility andperformance.


RAID Technologies

RAID - Redundant Array of Independent Disks

RAID technology provides data protection at the disk drive level. With RAID 1and RAID 5 technologies, access to the data is continuous even if a diskdrive fails.

RAID technologies available with Teradata:

RAID 1 Disk mirroring, used with both LSI Logic and EMC2 DiskArrays.

RAID 1+0 Disk mirroring with data striping, used with LSI Disk Arrays.Not needed with Teradata.

RAID 5 Data parity protection, interleaved parity, used with LSI LogicDisk Arrays.



RAID 1 – MirroringRAID 1 is data mirroring protection. The RAID 1 technology requires each primary datadisk to have a companion disk or mirror. The contents of the primary disk and the mirrordisk are identical.

When data is written on the primary disk, a write also occurs on the mirror disk. Themirroring process is invisible to the user. For this reason, RAID 1 is also called transparentmirroring.

With NCR RAID solutions, mirroring is managed by the controller, which provides a higherlevel of performance. Performance is improved because data can be read from either theprimary (data) drive or the mirror. The controller decides which read/write assembly (driveactuator) is closest to the requested data.

If the primary data disk fails, the mirror disk can be accessed without data loss. There is aminor performance penalty if a drive fails because the array controller can read from eitherdrive if both drives are available. If either disk fails, the disk array controller can copy thedata from the remaining drive to a replacement drive while normal operations continue.

RAID 1 – Striped MirroringWhen user data is to be written to the array, the controller instructs the array to write a blockof data to one drive pair to the defined stripe depth. Subsequent data blocks are writtenconcurrently to contiguous sectors in the next drive pair to the defined stripe depth. In thismanner, data are striped across the array of drives, utilizing multiple drives and actuators.

The example on the facing page illustrates RAID 1 Striped Mirroring. This RAIDtechnology is also referred to as RAID 1 + 0, RAID 0 + 1, or RAID 10.

With LSI Logic (previously, Symbios) arrays, striped mirroring is automatic when youcreate a drive group (with RAID 1 technology) that has multiple mirrored pairs of disks.

If an application (e.g., Teradata Database) uniformly distributes data, striped mirroring(RAID 1+0) and mirroring (RAID 1) will have similar performance.

If an application (database) partitions data, striped mirroring (RAID 1+0) can lead toperformance gains over mirroring (RAID 1) because array controllers equally spread I/O’sbetween channels in the array.

SSttrriippeedd MMiirrrroorriinngg iiss NNOOTT nneecceessssaarryy wwiitthh TTeerraaddaattaa..


RAID 1 – Mirroring

LUN 1LUN 0

Block A0 Block A1

Block A2 Block A3

Block A0

Block A1

Block A2

Block A3

Disk Array Controller

Block B0 Block B1

Block B2 Block B3

Block B0

Block B1

Block B2

Block B3

Mirror 3Disk 3Mirror 1Disk 1

• 2 Drive Groups each with 1 mirrored pair of disks• Operating system sees 2 logical disks (LUNs) or volumes• If LUN 0 has more activity , more disk I/Os occur on the first two drives in

the array.

2 Drive Groups -each with 1 pair ofmirrored disks

If physical drives are 36 GB each, then each logical unit(LUN) or volume is effectively 36 GB.



RAID 1 SummaryRAID 1 characteristics include:

• Data is fully replicated• Easy to understand technology• Follows a traditional approach• Transparent to the operating system• Redundant drive is affected only by write operations

RAID 1 advantages include:

• High I/O rate (small logical block size)• Maximum data availability• Minor performance penalty with single drive failure• No performance penalty in write intensive environments

RAID 1 disadvantages include:

• Only 50% of total disk space is available for user data. Therefore, RAID 1 has50% overhead in disk space usage.

Summary

• RAID 1 provides high data availability and performance, but storage costs are high.• Striped mirroring is not necessary with Teradata.


RAID 1 Summary

Characteristics

• data is fully replicated• striped mirroring is possible with multiple pairs of disks in a drive group• transparent to operating system

Advantages

• maximum data availability• read performance gains• no performance penalty with write operations• fast recovery and restoration

Disadvantages

• 50% of disk space is used for mirrored data

Summary

• RAID 1 provides high data availability and performance, but storage costsare higher.

•• Striped Mirroring is NOT necessary with Teradata.Striped Mirroring is NOT necessary with Teradata.



RAID 5 – Data Parity ProtectionWith the RAID 5 technology, user information and parity are combined on every disk in thearray. Independent and/or parallel data read and write operations are performed.

Data is spread across the disks one segment at a time. Parity is also striped across all disksin an interleaved fashion. Notice in the diagram on the following page how the paritysectors are located on different sectors on each disk. This reduces the parity updatebottleneck.

All write operations require a read-modify-write:

1. Read old data

2. Read old (existing) parity data

3. XOR old data + new data + old parity = new parity

4. Write new data

5. Write new parity data

Parity is effectively derived from an Exclusive-OR algorithm:

Data 1 + Data 2 + Data 3 = Parity

If a disk fails in RAID 5, the controller recovers the missing information by using the valueof the known data to calculate the missing information.


RAID 5 – Data Parity Protection


Disk 3Disk 1 Disk 2 Disk 4

Block 0 Block 1 Block 2 Parity

Block 3 Block 4 Parity Block 5

Block 6 Parity Block 7 Block 8

Parity Block 9 Block 10 Block 11

Block 12 Block 13 Block 14 Parity

LUN 0

• Sometimes referred to as “3 + 1” RAID 5.

• When data is updated, parity is also updated.

new_data XOR current_data XOR current_parity = new_parity

If physical drives are 36 GB each, then each logical unit(LUN) or volume is effectively 108 GB.



RAID 5 SummaryRAID 5 characteristics include:

• Data is striped across multiple disks dependent on stripe depth

• Data is distributed across all disks

• Parity is interleaved across all disks

• Independent actuators

• Transparent to system software

RAID 5 advantages include:

• Suitable for transaction processing

• Minimal disk space overhead – 20% or 25%

• High I/O rate in read intensive environments


RAID 5 Summary

Characteristics• data and parity is striped and interleaved across multiple disks• XOR logic is used to calculate parity• transparent to operating system

Advantages• provides high availability with minimum disk space (e.g., 25%) used for

parity overhead

Disadvantages• write performance penalty• performance degradation during data recovery and reconstruction

Summary– High data availability with minimum storage cost– Good choice when majority of I/O’s are reads and storage space is at a

premium



Teradata – RAID 1 and RAID 5 The facing page compares the use of RAID 1 and RAID 5 with Teradata.

The advantages of mirroring include:

Superior Performance

• Mirroring provides the best read and write throughput.• Maximizes the performance capabilities of controllers and disk drives. • Best performance when a drive has failed.• Less reconstruction impact when a drive has failed.

Superior Availability

• Less susceptible to a double disk failure in a RAID drive group.• Faster reconstruction of a failed drive - shorter vulnerability period during

reconstruction.

Superior Price/Performance - the performance advantage of RAID 1 outweighs theadditional cost for typical Teradata warehouses.


Teradata – RAID 1 and RAID 5

RAID 1 for TeradataMost useful with typical Teradata data warehouses (e.g., Active DataWarehouses).

RAID 5 for TeradataMost useful when creating archival data warehouses that require lessexpensive storage and where performance is not as important.

Why?

RAID 1 provides Superior Performance• Mirroring provides the best read and write throughput.• Maximizes the performance capabilities of controllers and disk drives.• Best performance when a drive has failed.• Less reconstruction impact when a drive has failed.

RAID 1 provides Superior Availability• Less susceptible to a double disk failure in a RAID drive group.• Faster reconstruction of a failed drive - shorter vulnerability period during

reconstruction.



CliquesA clique is a set of Teradata nodes that share a common set of disk arrays. In the event ofnode failure, all vprocs can migrate to another available node in the clique. All nodes in theclique must have access to the same disk arrays.

The illustration on the facing page shows a four-node clique. In this environment when all 4nodes are available, 7 AMP vprocs will execute on each node.

In the event of node failing, the remaining nodes would attempt to absorb all vprocs fromthe failed node.

Acronyms:

DAC – Disk Array Controller


Cliques


0 4 36…….

SMP001-4 AMPs

1 5 37…….

SMP001-5 AMPs

2 6 38…….

SMP002-4 AMPs

3 7 39…….

SMP002-5 AMPs

Clique – a set of SMPs that share a common set of disk arrays.



Teradata Vproc MigrationIf an SMP node (running Teradata) fails, Teradata restarts and the AMP vprocsthat were executing on the failed node are started on other nodes within theclique.

PE vprocs that are assigned to channel connections do not migrate to anotherSMP. PE vprocs that are assigned to gateway connections may or may not(depending on configuration) migrate to another node within the clique.


Teradata Vproc Migration

Vproc Migration – vprocs in the failed node are started in the remainingnodes within the “clique”.

SMP Fails


SMP001-4 AMPs

0 3 39…

SMP001-5 AMPs

1 4 37…….

SMP002-4 AMPs

2 5 38…….

SMP002-5 AMPs

36



LocksLocking prevents multiple users who are trying to change the same data at the same timefrom violating the data's integrity. This concurrency control is implemented by locking thedesired data. Locks are automatically acquired during the processing of a request andreleased at the termination of the request. In addition, users can specify locks.

There are four types of locks: Exclusive, Write, Read, and Access.

Exclusive locks are only applied to databases or tables, never to rows. They are the mostrestrictive type of lock; all other users are locked out. Exclusive locks are used rarely, mostoften when structural changes are being made to the database.

Write locks enable users to modify data while locking out all other users except readers notconcerned about data consistency (Access lock readers). Until a Write lock is released, nonew read or write locks are allowed.

Read locks are used to ensure consistency during read operations. Several users may holdconcurrent read locks on the same data, during which no modification of the data ispermitted.

Access locks can be specified by users who are not concerned about data consistency. Theuse of an access lock allows for reading data while modifications are in process. Accesslocks are designed for decision support on large tables that are updated only by small single-row changes. Access locks are sometimes called “stale read” locks, i.e. you may get ‘staledata’ that hasn’t been updated.

Three levels of database locking are provided:

• Database - locks all objects in the database• Table - locks all rows in the table or view• Row Hash - locks all rows with the same row hash

The type and level of locks are automatically chosen based on the type of SQL commandissued. The user has, in some cases, the ability to upgrade or downgrade the lock.

For example, if an SQL UPDATE command is executed without a WHERE clause, aWRITE lock is placed on the table. If an SQL UPDATE command is executed with aWHERE clause that specifies a Primary Index value, then a row hash lock is used.


Locks

Exclusive – prevents any other type of concurrent accessWrite – prevents other reads, writes, exclusivesRead – prevents writes and exclusivesAccess – prevents exclusive only

There are four types of locks:

Database – applies to all tables/views in the databaseTable/View – applies to all rows in the table/viewsRow Hash – applies to all rows with same row hash

Locks may be applied at three levels:

Lock types are automatically applied based on the SQL command:SELECT – applies a Read lockUPDATE – applies a Write lockCREATE TABLE – applies an Exclusive lock



Locking Modifier

This option precedes an SQL statement and locks a database, table, view, or row hash. Thelocking modifier overrides the default usage lock that Teradata places on a database, table,view, or row hash in response to a request.

ACCESSAccess locks have many advantages. This allows quick access to data, even if otherrequests are updating the data. They also have minimal effect on locking out others - whenyou use an access lock, virtually all requests are compatible with yours.

NOWAITIf a resource is locked and an application does not want to wait for that lock to be released,the Locking Modifier NOWAIT option can be used. The NOWAIT option indicates that ifthe lock cannot be obtained, then the statement will be aborted.

This option is used in situations where it is not desirable to have a statement wait forresources, possibly also tying up resources in the process of waiting.

Example:

LOCKING ROW FOR WRITE NOWAIT UPDATE ….. ;


Locking Modifier

LOCKING ROW FOR ACCESS SELECT * FROM TABLE_A;An “Access Lock” allows the user to access (read) an object that has a READ orWRITE lock associated with it.

In this example, even though an access row lock was requested, a table levelaccess lock will be issued because the SELECT causes a full table scan.

LOCKING ROW FOR EXCLUSIVE UPDATE TABLE_B SET A = 2002;This request asks for an exclusive lock, effectively upgrading the lock.

LOCKING ROW FOR WRITE NOWAIT UPDATE TABLE_C SET A = 2003;The locking with the NOWAIT option is used if you do not want your transaction towait in a queue.

NOWAIT effectively says to abort the the transaction if the locking manager cannotimmediately place the necessary lock.

The locking modifier overrides the default usage lock that Teradata places on adatabase, table, view, or row hash in response to a request.

Certain locks can be upgraded or downgraded:



Rules of LockingAs the facing page illustrates, a new lock request must wait (queue) behind otherincompatible locks that are either in queue or in effect. The new Read lock must wait untilthe write lock ahead of it is released before it goes into effect.

In the second example, the second Read lock request may occupy the same position in thequeue as the Read lock that was already there. When the current Write lock is released, bothrequests may be given access concurrently. This only happens when locks are compatible.

When an SQL statement provides row hash information, a row hash lock will be used. Ifmultiple row hashes within the table are affected, a table lock is used.


Rules of Locking

Lock requests are queuedbehind all outstandingincompatible lock requestsfor the same object.

Rule

Example 1 – New READ lock request goes to the end of queue.

READ WRITE READ READ WRITE READ

New request New lock queueLock queue Current lock Current lock

Example 2 – New READ lock request shares slot in the queue.

READ READ

New request New lock queueLock queue Current lock Current lock

READ WRITE WRITE

READ

LOCK LEVEL HELDLOCKREQUEST

ACCESS

READ

WRITE

EXCLUSIVE

NONE ACCESS READ WRITE EXCLUSIVE

Granted

Granted Granted

GrantedGranted

Granted

Granted

Granted

Granted Granted Queued

QueuedQueued

Queued

Queued

Queued

Queued

QueuedQueued

Queued



Access LocksAccess locks have many advantages. This allows quick access to data, even if otherrequests are updating the data. They also have minimal effect on locking out others - whenyou use an access lock, virtually all requests are compatible with yours.

When doing large aggregations of numbers, it may be inconsequential if certain rows arebeing updated during the summation, particularly if one is only looking for approximatetotals. Access locks are ideal for this situation.

Looking at Example 3, what happens to the Write lock request when the Read lock goesaway? Looking at the chart, it will be “Granted” since Write and Access are consideredcompatible.


Access Locks

Lock requests are queuedbehind all outstandingincompatible lock requestsfor the same object.

Rule

Example 3 – New ACCESS lock request granted immediately.

ACCESS WRITE WRITE READ

New request New lock queueLock queue Current lock Current locks

ACCESS

READ

Advantages of Access LocksPermit quicker access to table in multi-user environment.Have minimal ‘blocking’ effect on other queries.Very useful for aggregating large numbers of rows.

Disadvantages of Access LocksMay produce erroneous results if during table maintenance.

LOCK LEVEL HELDLOCKREQUEST

ACCESS

READ

WRITE

EXCLUSIVE

NONE ACCESS READ WRITE EXCLUSIVE

Granted

Granted Granted

GrantedGranted

Granted

Granted

Granted

Granted Granted Queued

QueuedQueued

Queued

Queued

Queued

Queued

QueuedQueued

Queued



FallbackFallback protects your data by storing a second copy of each row of a table on an alternative“fallback AMP”. If an AMP fails, the system accesses the fallback rows to meet requests.Fallback provides AMP fault tolerance at the table level. With Fallback tables, if one AMPfails, all of the table data is still available. Users may continue to use Fallback tableswithout any loss of available data.

When a table is created, or any time after its creation, the user may specify whether or notthe system should keep a fallback copy. If Fallback is specified, it is automatic andtransparent to the user.

Fallback guarantees that the two copies of a row will always be on different AMPs.Therefore, if either AMP fails, the alternate row copy is still available on the other AMP.

Certainly there is a benefit to protecting your data. However, there are costs associated withthat benefit. They are: twice the disk space for storage, and twice the I/O for Inserts,Updates, and Deletes. (However, the Fallback option does not require any extra I/O forSELECT operations and the fallback I/O will be performed in parallel with the primary I/O.)

The benefits of Fallback include protecting your data from hardware (disk) failure,protecting your data from software (node) failure, automatic recovery and minimumrecovery time after repairs or fixes are complete.

A hardware (disk) or software (vproc) failure causes an AMP tobe taken off-line until the problem is corrected.

During this period, Fallback tables are fully available to users.

When the AMP is brought back on-line, the associated Vdisk isrefreshed to reflect any changes during the off-line period.


FallbackA Fallback table is fully available in the event of an unavailable AMP.

A Fallback row is a copy of a “Primary row” which is stored on a different AMP.

Benefits of Fallback• Permits access to table data during AMP off-line period.• Adds a level of data protection beyond disk array RAID.• Automatic restore of data changed during AMP off-line.• Critical for high availability applications.

Cost of Fallback• Twice the disk space for table storage.• Twice the I/O for Inserts, Updates and Deletes.

Loss of any twoAMPs in a clustercauses RDBMS tohalt!

Note:

Primaryrows

Fallbackrows

AMP

2 611

3 5 128

17

3 85 2 1 11 6 127

AMP AMP AMP



Fallback ClustersA cluster is a group of AMPs that act as a single fallback unit. Clustering has no effect onthe distribution of the Primary rows of a table. The Fallback row copy however, will alwaysgo to a different AMP in the same cluster.

Cluster sizes are set through a Teradata console utility, and may range from 2 to 16 AMPsper cluster (not all clusters in a system have to be the same size). The example shows an 8-AMP system set up in two clusters of 4-AMPs each.

Should an AMP fail, the primary and fallback row copies stored on that AMP cannot beaccessed. However, their alternate copies are available through the other AMPs in the samecluster.

The loss of an AMP in one cluster has no effect upon other clusters. It is possible to loseone AMP in each cluster and still have full access to all Fallback-protected table data. Ifthere are two AMP failures in the same cluster, the entire Teradata system halts.

While an AMP is down, the remaining AMPs in the cluster must do their own work plus thework of the down AMP. The larger the size of the cluster, the less noticeable the workloadincrease is within that cluster when one AMP fails. On the other hand, large cluster sizes aremore vulnerable to a second failure before recovery from the first failure is complete, whichhalts the entire Teradata system.

A small cluster size (e.g., 2 AMP cluster) reduces the chances of have 2 down AMPs in asingle cluster which would cause a non-operational configuration. However, very smallcluster sizes (e.g, 2 AMP cluster) impacts performance. Therefore, a typical cluster size of 4AMPs provides the best balance in terms of performance and availability.

In the example on the facing page, if you …

Lose AMP 3 from cluster 0 - AMPs 1, 2, and 4 experience 33% increase in workload

Lose AMP 6 from cluster - AMPs 5, 7, and 8 experience 33% increase in workload

Lose AMP 7 from cluster (assuming AMP 6 is still down) - System halts


Fallback Clusters

• A Fallback cluster is a defined set of AMPs across which fallback is implemented.• All Fallback rows for AMPs in a cluster must reside within the cluster.• Loss of one AMP in the cluster permits continued table access.• Loss of two AMPs in the cluster causes the RDBMS to halt.

Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14

AMP 2 AMP 3 AMP 4Cluster 0

34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78

Primaryrows

Fallbackrows

AMP 5 AMP 6 AMP 7 AMP 8Cluster 1

41 766

93 72 88

58 2093 88 452 17 7237

45 7 17 37 58 41 20 2 66



Fallback and RAID ProtectionRAID 1 mirroring and RAID 5 data parity protection provide protection in the event of adisk drive failure.

Fallback provides another level of data protection beyond disk mirroring or data parityprotection.

Examples of other failures that Fallback provides protection against include:

• Multiple drive failures in the same drive group

• An array is not available (e.g., both disk array controllers fail in the disk array)

• An AMP is not available (e.g., a software problem)


Fallback and RAID Protection

• RAID 1 Mirroring or RAID 5 Data Parity Protection provides protection in theevent of disk drive failure.

– Provides protection at a hardware level

– Teradata is unaware of the RAID technology used

• Fallback provides an additional level of data protection and provides accessto data when an AMP is not available (not online).

• Additional types of failures that Fallback protects against include:

– Multiple drives fail in the same drive group,

– Disk array is not available

• Both disk array controllers fail in a disk array

• Two of the three power supplies fail in a disk array

– AMP is not available (e.g., software or data error)



Fallback and RAID 1 ExampleThe next set of pages contains an example of how Fallback and RAID 1 Mirroring worktogether.


Fallback and RAID 1 Example

Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14

AMP 2 AMP 3 AMP 4Vdisk

34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78

RAID 1 -Mirrored Pair of PhysicalDisk Drives

Primary 342250

Fallback 1938

8

Primary 342250

Fallback 1938

8

Primary 141

38Fallback 50

2778

Primary 141

38Fallback 50

2778

Primary 628

27Fallback 5

3414

Primary 628

27Fallback 5

3414

Primary 57819

Fallback 2262

1

Primary 57819

Fallback 2262

1

This example assumes that RAID 1 Mirroring is used and the table is fallback protected.



Fallback and RAID 1 Example (cont.)The example of how Fallback and RAID 1 Mirroring work together is continued.

In this example, one disk drive has failed in the first drive group. Is Fallback needed? No.As a matter of fact, Teradata doesn’t even realize that the drive has failed. The disk arraycontinues to provide access to the data directly from the second disk drive in the drivegroup. The disk array controller will send a “fault” or error message to the AWS.


Fallback and RAID 1 Example (cont.)

Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14


34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78


Primary 342250

Fallback 1938

8

Primary 342250

Fallback 1938

8

Primary 141

38Fallback 50

2778

Primary 141

38Fallback 50

2778

Primary 628

27Fallback 5

3414

Primary 628

27Fallback 5

3414

Primary 57819

Fallback 2262

1

Primary 57819

Fallback 2262

1

Assume one disk drive fails. Is Fallback needed is this example?




In this example, assume two disk drives have failed – one in the first drive group and one inthe third drive group. Is Fallback needed? No. Like before, Teradata doesn’t even realizethat the drives have failed. The disk array continues to provide access to the data directlyfrom the second disk drive each of the drive groups. The disk array controller will send“fault” or error messages to the AWS.



Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14


34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78


Primary 342250

Fallback 1938

8

Primary 342250

Fallback 1938

8

Primary 141

38Fallback 50

2778

Primary 141

38Fallback 50

2778

Primary 628

27Fallback 5

3414

Primary 628

27Fallback 5

3414

Primary 57819

Fallback 2262

1

Primary 57819

Fallback 2262

1

Assume two disk drives have failed. Is Fallback needed is this example?




In this example, assume two disk drives have failed – both failed drives are in the first drivegroup. Is Fallback needed? Yes, if you need to access the data in this table. When multipledisk drives fail in a drive group, the data (Vdisk) is not available and the AMP goes into aFATAL state. At this point, Teradata does realize that an AMP is not available and Teradatarestarts. The disk array controller will send “fault” or error messages to the AWS.

The AWS will also get “fault” messages indicating that Teradata has restarted.




Primary 342250

Fallback 1938

8

Primary 342250

Fallback 1938

8

Primary 141

38Fallback 50

2778

Primary 141

38Fallback 50

2778

Primary 628

27Fallback 5

3414

Primary 628

27Fallback 5

3414

Primary 57819

Fallback 2262

1

Primary 57819

Fallback 2262

1

Assume two disk drives have failed in the same drive group. Is Fallback needed?

Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14


34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78




In this example, assume three disk drives have failed – two failed drives are in the first drivegroup and one failed drive is in the third drive group. Is Fallback needed? Yes, if you needto access the data in this table. When multiple disk drives fail in a drive group, the data(Vdisk) is not available and the AMP goes into a FATAL state. However, the third AMP isstill operational and online.



Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14


34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78


Primary 342250

Fallback 1938

8

Primary 342250

Fallback 1938

8

Primary 141

38Fallback 50

2778

Primary 141

38Fallback 50

2778

Primary 628

27Fallback 5

3414

Primary 628

27Fallback 5

3414

Primary 57819

Fallback 2262

1

Primary 57819

Fallback 2262

1

Assume three disk drives failures. Is Fallback needed? Is the data still available?



Fallback vs. non-Fallback Tables SummaryFallback tables have a major advantage in terms of availability and recoverability. Theycan withstand an AMP failure in each cluster and maintain full data availability. A secondAMP failure in any cluster results in a system halt. A manual restart of the system isrequired in this circumstance.

Non-Fallback tables are affected by the loss of any one AMP. The table continues to beaccessible, but only for those AMPs that are still on-line. A one-AMP Primary Index accessis possible, but a full table scan is not.

Fallback tables are easily recovered after a failure due to the availability of Fallback rows.Non-Fallback tables may only be restored from external medium in the event of a disaster.


Fallback vs. non-Fallback Tables Summary

FALLBACK TABLES

One AMP Down - Data fully available

Two or more AMPs Down

AMP AMP AMP AMP

- If different cluster,data fully available

- If same cluster,Teradata halts

AMP AMP AMP AMP

Non-FALLBACK TABLES

One AMP Down - Data partially available;queries that avoiddown AMP succeed.

Two or more AMPs Down

AMP AMP AMP AMP

- If different cluster,data partially available;queries that avoiddown AMP succeed.

- If same cluster,Teradata halts

AMP AMP AMP AMP



Clusters and CliquesAs your know, a cluster is a group of AMPs that act as a single fallback unit. A clique is aset of Teradata nodes that share a common set of disk arrays. Clusters provide data accessprotection in the event of an AMP failure (usually because of a Vdisk failure). Cliquesprovides protection from SMP node failures.

The best availability for Teradata is to spread clusters across different cliques. The “DefaultCluster” function of the CONFIG utility does this automatically.

The example on the facing page illustrates a 16-node system. Each clique is connected to aset of 4 disk arrays. This example assumes each SMP is configured with 10 AMPs.


Clusters and Cliques

0 4 36… 1 5 …... 2 6 …... 3 7 ....

SMP001-4 SMP001-5 SMP002-4 SMP002-5

39

40 44 …...41 45 …... 42 46 …... 43 47 ...…

SMP003-4 SMP003-5 SMP004-4 SMP004-5

80 84 …... 81 85 …... 82 86 …... 83 87 ...…

SMP005-4 SMP005-5 SMP006-4 SMP006-5

120 124 ...… 121 125 …...

SMP007-4 SMP007-5 SMP008-4 SMP008-5

122 126 …... 123 127 …...

160 Disks in MultipleDisk Arrays for Clique 0




Cluster 0 Cluster 1

Clique 0

Clique 1

Clique 2

Clique 3



Recovery Journal for Down AMPsAfter the loss of any AMP, a Down-AMP Recovery Journal is started automatically. Itspurpose is to log any changes to rows which reside on the down AMP. Any inserts, updates,or deletes affecting rows on the down AMP, are applied to the Fallback copy within thecluster. The AMP that holds the Fallback copy logs the Row ID in its Recovery Journal.

This process continues until such time as the down AMP is brought back on-line. As part ofrestart activity, the Recovery Journal is read and changed rows are applied to the recoveredAMP. When the journal has been exhausted, it is discarded and the AMP is brought on-linefully recovered.


Recovery Journal for Down AMPs

Automatically activated when an AMP is taken off-line.Maintained by other AMPs in the cluster.Totally transparent to users of the system.

Recovery Journal is:

While AMP is off-line Journal is active.Table updates continue as normal.Journal logs Row IDs of changed rows for down-AMP.

When AMP is back on-line Restores rows on recovered AMP to current status.Journal discarded when recovery complete.

Primaryrows

Fallbackrows

AMP 1

62 278

5 34 14


34 5022 5 1978 14 381

19 38 8 22 62 1 50 27 78

RecoveryJournal Row ID for 62Row ID for 34 Row ID for 14



Transient JournalThe Transient Journal exists to permit the successful rollback of a failed transaction.Transactions are not committed to the database until an End Transaction request has beenreceived by the AMPs, either implicitly or explicitly. Until that time, there is always thepossibility that the transaction may fail in which case the participating table(s) must berestored to their pre-transaction state.

The Transient Journal maintains a copy of all before images of all rows affected by thetransaction. If the event of transaction failure, the before images are reapplied to theaffected tables, the images are deleted from the journal and a rollback operation iscompleted. In the event of transaction success, at the point of transaction commit, the beforeimages for the transaction are discarded from the journal.


Transient Journal

Transient Journal – provides transaction integrity

• A journal of transaction “before images”.• Provides for automatic rollback in the event of TXN failure.• Is automatic and transparent.• “Before images” are reapplied to table if TXN fails.• “Before images” are discarded upon TXN completion.

BEGIN TRANSACTIONUPDATE Row A – Before image Row A recorded (Add $100 to checking)UPDATE Row B – Before image Row B recorded (Subtract $100 from savings)

END TRANSACTION – Discard before images

Successful TXN

BEGIN TRANSACTIONUPDATE Row A – Before image Row A recordedUPDATE Row B – Before image Row B recorded

(Failure occurs)(Rollback occurs) – Reapply before images(Terminate TXN) – Discard before images

Failed TXN



Permanent JournalThe purpose of the Permanent Journal is to provide selective or full database recovery to aspecified point in time. It permits recovery from unexpected hardware or software disasters.The Permanent Journal also has the effect of reducing the need for full table backups whichcan be costly both in time and resources.

The Permanent Journal is an optional journal and its features must be customized to thespecific needs of the installation. The journal may capture before images (for rollback),after images (for rollforward), or both. Additionally, the user must specify if single images(default) or dual images (for fault-tolerance) are to be captured.

A Permanent Journal may be shared by multiple tables or multiple databases.The journal captures images concurrently with standard table maintenance and queryactivity. The cost in additional required disk space may be calculated in advance to ensureadequate disk reserve.

The journal is periodically dumped to external media, thus reducing the need for full tablebackups - in effect, only the changes are backed up.


Permanent Journal

The Permanent Journal is an optional, user-specified, system-maintainedjournal which is used for recovery of a database to a specified point in time.

The Permanent Journal:• Is used for recovery from unexpected hardware or software disasters.• May be specified for ...

a.) One or more tablesb.) One or more databases

• Permits capture of Before Images for database rollback.• Permits capture of After Images for database rollforward.• Permits archiving change images during table maintenance.• Reduces need for full table backups.• Provides a means of recovering NO FALLBACK tables.• Requires additional disk space for change images.• Requires user intervention for archive and recovery activity.



Archiving and Recovering DataThe purpose of the ARC utility is to allow for the archiving and restoring of databaseobjects which may have been damaged or lost. There are several scenarios where restoringobjects from external media may be necessary.

• Restoring of non-Fallback tables after a disk failure.

• Restoring of tables which have been corrupted by batch processes which may haveleft the data in an ‘uncertain’ state.

• Restoring of tables, views or macros which have been accidentally dropped by theuser.

• Miscellaneous user errors resulting in damaged or lost database objects.

NCR’s Open Backup architecture provides two solutions from two NCR Partners.

• NetVault – from BakBone software

• NetBackup – from VERITAS Software (limited support)

The ASF2 utility is an older utility that provides an X Windows based front-end for creationand execution of ARC command scripts. It is designed to run on UNIX MP-RAS.


Archiving and Recovering Data

ARC

• The Archive/Restore utility (arcmain)• Runs on IBM, UNIX, and Windows 2000 systems• Archives and restores data from/to Teradata RDBMS• Restores or copies data from archive media• Permits data recovery to a specified checkpoint (using Permanent Journals)• ARC 7.0 is required to archive/restore with Teradata V2R5

Open Teradata Backup

• Two choices from different NCR Partners– NetVault - from BakBone software– NetBackup - from VERITAS software (limited support)

• Provides Windows front end for ARC• Easy creation of scripts for archive/recovery• Provides job scheduling and tape management functions• ASF2 no longer supported with Teradata V2R5


Review Questions

Match the item to a lettered description.

a.) Provides for TXN rollback in case of failureb.) Open Teradata Backup applicationc.) Protects all rows of a tabled.) Logs changed rows for down AMPe.) Provides for recovery to a point in time f.) Applies to all tables and views withing.) Multi-platform archive utilityh.) Lowest level of protection granularity i.) Protects tables from AMP failure j.) Protects database from a physical drive failurek.) Group of AMPs used by Fallback

____ 1.) Database locks____ 2.) Table locks____ 3.) Row Hash locks____ 4.) FALLBACK____ 5.) Cluster____ 6.) Recovery journal____ 7.) Transient journal____ 8.) ARC____ 9.) NetBackup/NetVault____ 10.) Permanent journal____ 11.) Disk Array



Notes

Introduction to NCR Systems Page 11-1

Module 11


Introduction to NCR Systems


• Specify 2 improvements of the 53xx systems as compared tothe 52xx systems.

• Define the purpose of the major subsystems that are part of aNCR MPP system.

• Describe the purpose of the various BYNET components.

• List the two LANs that connect the AWS to a 49xx/53xx systemand briefly describe the purpose of each LAN.

• Specify the names of the SMPs in an SMP cabinet.


Page 11-2 Introduction to NCR Systems

Notes


Table of Contents

NCR SYSTEMS ..................................................................................................................................................... 11-4SMP ARCHITECTURE ....................................................................................................................................... 11-6NCR CABINET PICTURES ................................................................................................................................ 11-8NCR 4800/4850 AND 5200/5250 SYSTEMS...................................................................................................... 11-10

NCR 4800 AND 5200 SYSTEMS ......................................................................................................................... 11-10NCR 4850 AND 5250 SYSTEMS ......................................................................................................................... 11-10

NCR 4851/4855 AND 5251/5255 SYSTEMS...................................................................................................... 11-12NCR 4851/5251 SYSTEMS................................................................................................................................. 11-12NCR 4855/5255 SYSTEMS................................................................................................................................. 11-12

NCR 4900 AND 5300 SYSTEMS........................................................................................................................ 11-14NCR 4950/4980 AND 5350/5380 SYSTEMS...................................................................................................... 11-16NCR 4475 AND 4480 SERVERS........................................................................................................................ 11-18MAKING SENSE OF THE DIFFERENT PLATFORMS............................................................................... 11-20COMPARING PERFORMANCE OF NCR SERVERS .................................................................................. 11-22NCR COEXISTENCE COMBINATIONS........................................................................................................ 11-24WHAT IS THE BYNET™?................................................................................................................................ 11-26BYNET 4 AND BYNET 16 SWITCHES ........................................................................................................... 11-28

4700/5150 BYNET SWITCHES (NOT SHOWN).................................................................................................... 11-28BYNET 64 SWITCHES ...................................................................................................................................... 11-30

BYNET V2 SWITCH CABINETS ......................................................................................................................... 11-30BYNET EXPANSION SWITCHES (BYB64G) ................................................................................................ 11-32BYNET EXPANSION SWITCHES (CONT.) ................................................................................................... 11-34BYNET™ SOFTWARE...................................................................................................................................... 11-36

BYNET™ DEVICE DRIVERS ............................................................................................................................. 11-36BLM.............................................................................................................................................................. 11-36BDL .............................................................................................................................................................. 11-36

ADMINISTRATION WORKSTATION (AWS)............................................................................................... 11-38ADDITIONAL AWS CAPABILITIES...................................................................................................................... 11-38

Dual or Multiple AWS .................................................................................................................................. 11-38Remote AWS ................................................................................................................................................. 11-38

AWS CONNECTIVITY TO A SYSTEM.......................................................................................................... 11-40SYSTEM LAN (SLAN)....................................................................................................................................... 11-40PRIVATE LAN (PVTLAN) ................................................................................................................................. 11-40

AWS NAMING CONVENTIONS...................................................................................................................... 11-42AWS SYSTEM DISPLAY (UNIX)..................................................................................................................... 11-44

AWS SYSTEM WINDOW .................................................................................................................................... 11-44CSF FAULTS WINDOW....................................................................................................................................... 11-44

AWS PROCESSOR LEVEL DISPLAY (UNIX) .............................................................................................. 11-46WINDOWS 2000 AWS EXAMPLE................................................................................................................... 11-48SUMMARY .......................................................................................................................................................... 11-50REVIEW QUESTIONS....................................................................................................................................... 11-52



NCR SystemsAs the competitive needs of businesses change, so should the system architecturechange. To be best-in-class, an information processing system in today'senvironment will typically have the following characteristics.

• Utilization of multiple processors to achieve acceptable performance inhandling many and/or complex transactions. Typically, this requiresparallel and/or distributed processing capabilities.

• Be capable of handling a very large database of current business

information, rapidly process complex queries, maintain data security, andbe accessible to the total enterprise.

• Be designed of an open architecture and implement a client/server

environment that is capable of integrating with multi-vendor products.

• Support on-line transaction processing as well as decision supportapplications.

• Provide access to an extensive set of CASE development tools and 5th

Generation Language products.

• Easily scalable in both processing power and data storage capacity withadherence to all industry-standard interfaces.

• In today’s global and highly competitive markets, computing systems

(especially enterprise servers) need to be available to the world 24 hours aday.

NCR’s Systems meet these requirements. NCR Systems prior to the NCR4950/5350 were also referred to as NCR WorldMark Systems.


NCR Systems

This module will discuss the following NCR systems:1 – 4 Nodes 2 – 512 NodesNCR 4800/4850 NCR 5200/5250NCR 4851/4855 NCR 5251/5255NCR 4900 NCR 5300NCR 4950 NCR 5350NCR 4980 NCR 5380

The basic building block is the SMP (Symmetric Multi-processing) node.

Common characteristics of these systems:

• MPP systems that use the BYNET interconnect• Support of the Teradata database – Version 2• Single point of operational control – AWS• Rack-based systems – each technology is encapsulated in its own chassis

Key differences:

• Speed and capacity of SMP nodes and systems• Cabinet architecture• BYNET interface cards, switches and speeds



SMP ArchitectureThe SMP or “processing node” is the basic building block for NCR systems. Theprocessing node contains the primary processor logic (CPUs), memory, and I/Ofunctionality.


SMP Architecture

SMP (Symmetrical Multi-Processing) Node - Basic building block of NCRsystems.

• Multiple CPUs• Shared memory• Under control of single O.S.

Other names include node, compute node, processing node, SMP node, 2-waynode, 4-way node, or UNIX (Windows) node.

Key hardware components are:

• CPUs and cache memory• Memory• System Bus• I/O Subsystem

System Bus

CPUs (2 or 4)

CPU

CPU

CPU

CPU

Memory

Memory

Memory

High Performance

PQSor

Fibre Channel Adapter

SCSI or Fibre C

hannel

I/O Subsystem



NCR Cabinet PicturesThe facing page has pictures of the NCR 4475 (Pedestal) and the NCR 5xxx cabinet or rack.

NCR 48xx, 49xx, 52xx, and 53xx systems use an industry standard rack mount architectureand individual chassis that conform to industry standards. Types of chassis that can beplaced in a rack or cabinet include:

• Processing Node• BYNET V2 16 Node Switch (BYA16G)• Server Management• Uninterruptable Power Supply

The rack is referred to as a 40U rack. A U represents a “unit of vertical measurement” in anindustry standard rack.

1U = 4.445 cm or 1.75” high

Therefore, this cabinet has 40U or 40 x 1.75” (70”) of usable space.


NCR Cabinet Pictures

NCR 4475(Pedestal)

1 SMP

NCR 4xxx/5xxx(Rack)

Can hold up to4 SMPs

Example of 5380SMP Chassis



NCR 4800/4850 and 5200/5250 SystemsThe NCR 4800/4850 and 5200/5250 systems utilize a rack-based cabinet. Like other NCRsystems, the 4800/4850 and 5200/5250 systems are composed of different type ofsubsystems.

The processing node is housed in an 11U “chassis” which is mounted in rack-based cabinet.A 4800/4850 or 5200/5250 cabinet is capable of housing two processing nodes.

NCR 4800 and 5200 SystemsThe key component is the SMP node – it is based on the Intel 100 MHz internal busarchitecture, uses the Intel Pentium II Xeon 450 MHz or Intel Pentium III Xeon500 or 550 MHz CPUs, and has 3 PCI buses. Simply stated, the 4800/5200 SMP is a fastercomputing engine than the previous 4700/5150.

NCR 4800 and 5200 systems can be upgraded to 4850 and 5250 systems.

NCR 4850 and 5250 SystemsThe NCR 4850 and 5250 systems are very similar to the 4800 and 5200 systems. The keydifference is that 4850 and 5250 SMPs utilize the Intel Pentium III Xeon 700 MHzCPU.

The 4850/5250 SMP chassis is 11U in height, same height as the 4800/5200 SMP chassis.

The following table lists the height of each chassis:

Chassis HeightServer Management 3U (13.3 cm, 5.25 in)BYA16G 3U (13.3 cm, 5.25 in)SMP – 4800/4850/5200/5250 11U (48.9 cm, 19.25 in.)UPS – 4800/4850/5200/5250 3U (13.3 cm, 5.25 in.)


NCR 4800/4850 and 5200/5250 Systems

Front View with PanelsFront View

UPS 3

UPS 3

UPS 3

SMP - 4 way node

SMP - 4 way node

BYNET V2

BYNET V2

SMC

2 SMP Nodes (11U)

4800/5200 4 Intel Xeon CPUs

(450/500/550 MHz)

4850/5250 4 Intel Xeon CPUs

(700 MHz)

1 - 4 GB Memory 3 PCI Buses 11 PCI slots

Example of BYNET V2(BYNET 16 switch)

Server Management (CMIC2)

Three 3U UPSwith Dual AC

Notes:

• 4800/4850 – up to 4 SMPsand uses BYNET 4 PCIboard switches.

• 5200/5250 – up to 512SMPs and uses BYNETchassis switches.

• SMPs and components arehoused in chassis moduleswhich are mounted in arack-based cabinet.

• SMPs use Intel® Pentium®III XeonTM CPUs, a 100MHz system bus, andleverage 4400 SMPtechnology.



NCR 4851/4855 and 5251/5255 SystemsThe NCR 4851/4855 and 5251/5255 systems have a different SMP architecture than the4800/4850/5200/5250 SMPs. The module chassis for these SMPs is 7U in height ascompared to 11U for the NCR 4800/4850/5200/5250 SMPs.

NCR 4851/5251 SystemsThe NCR 4851 and 5251 SMPs utilize the Intel Pentium III Xeon 700 MHz CPU.The 4851 and 5251 systems will primarily be used in a co-existence environment withexisting 4850 and 5250 systems.

NCR 4855/5255 SystemsOne key performance difference is that 4855 and 5255 SMPs utilize the Intel Pentium IIIXeon 900 MHz CPU. This faster CPU provides approximately a 15% performance gain.


Chassis HeightService Management 3U (13.3 cm, 5.25 in)BYA16G 3U (13.3 cm, 5.25 in)SMP – 4851/4855/5251/5255 7U (31.1 cm, 12.25 in.)UPS – 4851/4855/5251/5255 3U (13.3 cm, 5.25 in.)


NCR 4851/4855 and 5251/5255 Systems

Notes:

• 4851/4855 – up to 4 SMPs;use BYNET 4 V2 switches.

• 5251/5255 – up to 512SMPs; use BYNET V2switches.

• SMPs are based on the4455 SMP architectureand are housed in achassis that is 7U inheight.

• 4855/5255 SMPs useIntel® Pentium® III 900XeonTM MHz CPUs, a 100MHz system bus, andleverage 4455 SMPtechnology (4851/5251SMPs use 700 MHz CPUs).

Front View with Panels

2 SMP Nodes (7U)

4 Intel CPUs(Xeon 700/900 MHz)

1 - 4 GB Memory3 PCI Buses 8 PCI Slots

BYNET 16 switches (Optional)



Front View

UPS 3

UPS 3

UPS 3

SMP - 4 way node

4U Space

SMP - 4 way node

4U Space

BYNET V2

BYNET V2

SMC



NCR 4900 and 5300 SystemsThe NCR 4900 and 5300 systems have a new SMP architecture. The module chassis forthese new SMPs is 5U in height as compared to 7U for the 4851/4855/5251/5255 SMPs.

The NCR 4900 and 5300 SMPs utilize the Intel Pentium III Tualatin 1.4 GHz CPU.

One key difference with this generation is that a cabinet or rack can now hold 4 SMPs ratherthan two.


Chassis HeightService Management 3U (13.3 cm, 5.25 in)BYA16G 3U (13.3 cm, 5.25 in)SMP – 4900/5300 5U (22.2 cm, 8.75 in.)UPS – 4900/5300 2U (8.9 cm, 3.5 in.)


NCR 4900 and 5300 Systems

Front View with Panels

4 SMP Nodes (5U)

2 Intel CPUs(Pentium III 1.4 GHz)

1 - 4 GB Memory3 PCI Buses

6 PCI Slots

BYNET 16 switches (Optional)



Front View

SMP 2 CPUs

SMP 2 CPUs

SMP 2 CPUs

SMP 2 CPUs

BYNET V2

BYNET V2

SMC

4U Space

1U Space

Notes:

• 4900 – up to 4 SMPs; usesBYNET 4 V2.1 switch.

• 5300 – up to 512 SMPs;use BYNET V2 switches.

• SMPs are housed in achassis that is 5U inheight and this allows for4 SMPs in the rack.

• Each 4900/5300 SMP hastwo Intel® Pentium® III 1.4GHz CPUs and a 133 MHzsystem bus.



NCR 4950/4980 and 5350/5380 SystemsThe NCR 4950/4980 and 5350/5380 system platforms use industry standard rack mountarchitecture. It consists of individual subsystem chassis that are housed in standard rackframes. Subsystems are self-contained, and their configurations — either internal or withina system — are redundant. The design ensures overall system reliability, enhances itsserviceability, and enables time and cost efficient upgrades.

The NCR 4950/4980 and 5350/5380 rack-based cabinets house up to 4 SMPs. The modulechassis for these SMPs is 5U in height (same height as the 4900/5300 SMP.

Each 4950 and 5350SMP has two Intel Pentium IV Xeon 2.8 GHz CPUs.

Each 4980 and 5380SMP has two Intel Pentium IV Xeon 3.06 GHz CPUs.

NCR 4950/5350 and 4980/5380 systems are composed of different type of subsystems. Themost important subsystem is the SMP (processing node) which is the basic building block ofthe system. The processing node is implemented in a “chassis” which is mounted in therack-based cabinet. The 4950/5350 and 4980/5380 cabinets/racks are capable of housing upto four processing nodes.


Chassis HeightService Management 3U (13.3 cm, 5.25 in)BYA16G 3U (13.3 cm, 5.25 in)SMP – 4950/5350 and 4980/5380 5U (22.2 cm, 8.75 in.)UPS – 4950/5350 and 4980/5380 2U (8.9 cm, 3.5 in.)


NCR 4950/4980 and 5350/5380 Systems


SMPxxx-4

BYNET V2

BYNET V2

SMC

2U Space

1U Space

1U Space

1U Space

SMPxxx-5

SMPxxx-6

SMPxxx-7

4 SMP Nodes (5U)

2 Intel CPUs(Pentium IV 2.8 or3.06 GHz)

1 - 4 GB Memory3 PCI Buses

6 PCI Slots

BYNET 16 switches1U or 3U

(Optional)



Notes:

• 4950/4980 – up to 4 SMPs;uses BYNET 4 V2.1 switch.

• 5350/5380 – up to 512SMPs; use BYNET V2switches.

• SMPs are housed in achassis that is 5U in heightand this allows for 4 SMPsin the rack.

• Each 4950/5350 SMP hastwo Intel® Pentium® IV 2.8GHz Xeon CPUs and a 400MHz system bus (FSB).

• Each 4980/5380 SMP hastwo Intel® Pentium® IV3.06 GHz Xeon CPUs and a533 MHz system bus(FSB).



NCR 4475 and 4480 ServersThe NCR 4475 or 4480 server is an entry-level data warehousing system server. Whencombined with the Teradata database, external storage, connectivity features andapplications, the NCR 4475/4480 provides a complete entry-level data warehousingenvironment.

The NCR 4475/4480 is an integrated system designed to be a complete solution for an entry-level database. The system can run either the UNIX MP-RAS or Microsoft Windows 2000operating system.

The base components of the NCR 4475 server are the same as the 4950/5350 processingnode. Characteristics include:

• Processor and Bus Speed– Pentium 4 XEON CPUs, 2.8 GHz/512KB cache each – 400 MHz front side bus (FSB)

The base components of the NCR 4480 server are the same as the 4980/5380 processingnode. Characteristics include:

• Processor and Bus Speed– Pentium 4 XEON CPUs, 3.06 GHz/512KB cache each – 533 MHz front side bus (FSB)

Characteristics of both include:

• Memory 1 GB memory minimum– 4 GB max. for MP-RAS Teradata or non-Teradata– 4 GB max. for Windows Teradata– 6 GB max. for Windows non-Teradata

• I/O Slots - 6 PCI slots:– Two 64-bit/100 MHz slots (PCI Bus 1, slots 1, 2)– Three 32-bit/33 MHz slots (PCI Bus 0, slots 3, 4, and 5)– One 64-bit/133 MHz slot (PCI Bus 2, slot 6)

• Drive Bays – 1 flex drive (3.5")– 2 removable media bays (5.25”) populated with a CD-ROM drive and a tape

drive– 5 hot-pluggable disk bays housing four 18 GB hard drives standard and an

optional fifth 18 GB or 36 GB hard drive


NCR 4475 and 4480 Servers

NCR 4475 and 4480 Servers

• NCR 4475 and 4480 are entry-level data warehousing system servers

• NCR 4475 has similar architecture to 4950/5350 SMP (same baseboard)

– Utilizes one or two Intel Pentium IV 2.8 GHz CPUs; 400 MHz FSB– Up to 4 GB of memory as a Teradata server– Multiple fast (133 MHz) and wide (64-bit PCI buses)

• NCR 4480 has similar architecture to 4980/5380 SMP (same baseboard)

– Utilizes one or two Intel Pentium IV 3.06 GHz CPUs; 533 MHz FSB– Up to 4 GB of memory as a Teradata server– Multiple fast (133 MHz) and wide (64-bit PCI buses)

• Designed for pedestal or deskside environments

• External Disk Arrays - deskside

– MP-RAS - NCR 6288– Windows 2000 - NCR 6283 and 6284



Making Sense of the Different PlatformsThe facing page attempts to provide some perspective of the different platforms.

Notes:

Xeon – Intel Pentium III Xeon CPUP III – Intel Pentium III Xeon CPU

The 4400, 4800, 4850, 5200, and 5250 SMPs are based on the Intel Aspen baseboardtechnology. These nodes may be referred to as Aspen nodes.

The 4455, 4851, 4855, 5251, and 5255 SMPs are based on the Intel Koa baseboardtechnology. These nodes may be referred to as Koa nodes.

The 4470, 4900 and 5300 SMPs are based on the INTEL Dodson baseboard technology andmay be referred to as Dodson nodes.

The 4475, 4950 and 5350 SMPs are based on the INTEL Hodges baseboard technology andmay be referred to as Hodges nodes.

The 4480, 4980, and 5380 SMPs are based on the INTEL Harlingen baseboard technologyand may be referred to as Harlingen nodes.

The following dates indicate when these systems were generally available to customers(GCA – General Customer Availability).

- 5100M January, 1996 (not described in this course)- 4700/5150 January, 1998 (not described in this course)- 4800/5200 April, 1999- 4850/5250 June, 2000- 4851/4855/5251/5255 July, 2001- 4900/5300 March, 2002- 4950/5350 December, 2002- 4980/5380 August, 2003


Making Sense of the Different Platforms

SMP MPP

4400Xeon 450 -500 - 550 MHz

4800 (2 - 4 SMPs)Xeon 450/500/550 MHzBYNET V1.1

4850 (2 - 4 SMPs)Pentium III Xeon 700 MHzBYNET V2

5250 (2 - 512 SMPs)Pentium III Xeon 700 MHzBYNET V2

4851/4855 (2 - 4 SMPs)P III Xeon 700/900 MHzBYNET V2

5251/5255 (2 - 512 SMPs)P III Xeon 700/900 MHzBYNET V2

1999

2000

2001

2002

2003

4455Xeon 700 MHz

5200 (2 - 512 SMPs)Xeon 450/500/550 MHzBYNET V2

4900 (1 - 4 SMPs)Pentium III 1.4 GHzBYNET V2.1

4470Pentium III1.4 GHz

5300 (2 - 512 SMPs)Pentium III 1.4 GHzBYNET V2.1

4950/4980 (1 - 4 SMPs)Pent. IV Xeon 2.8/3.06 GHzBYNET V2.1

4475/4480Pentium Xeon IV2.8/3.06 GHz

5350/5380 (2 - 512 SMPs)Pent. IV Xeon 2.8/3.06 GHzBYNET V2.1

Aspen

Koa

Dodson

Hodges

Harlingen



Comparing Performance of NCR ServersThe facing page attempts to provide some perspective as to the relative performancebetween 52xx systems and 53xx systems.


Comparing Performance of NCR Servers

52554 CPUs900 MHz

52554 CPUs900 MHz

1.15

1.15

2.3

5255

1.0

5250/5251

4 CPUs700 MHz

5250/52514 CPUs700 MHz

1.0

2.0

5250/5251

53002 CPUs1.4 GHz

53002 CPUs1.4 GHz

53002 CPUs1.4 GHz

53002 CPUs1.4 GHz

.86

3.44

.86

.86

.86

5300

RelativePerformance

Notes: Total number of CPUs per rack in each of these configurations is 8.Hyper-threading with 5380 will yield performance numbers higher than 1.30.

5350/53802 CPUs

2.8/3.0 GHz

5350/53802 CPUs

2.8/3.0 GHz

5350/53802 CPUs

2.8/3.0 GHz

5350/53802 CPUs

2.8/3.0 GHz1.22 / 1.30+

4.88 / 5.20+

1.22 / 1.30+

1.22 / 1.30+

1.22 / 1.30+

5350/5380



NCR Coexistence CombinationsNCR provides investment protection by allowing coexistence of different platforms in asingle system. Different options are available to the customer - upgrades, expansion,migration, coexistence, and coresidence. Coexistence and coresidence can both be viewedas types of expansion. Coexistence systems and coresidence systems both contain a mixture of nodes and storagethat operate as a single MPP system running the same software. Each system is managedwith the same AWS.

The general rule is that 2 generations of systems can coexist. However, there are situationswhere 3 generations of systems can coexist. For example, it is possible to have a systemconfiguration with 5255 SMPs, 5300 SMPs, and 5350 SMPs.

There are basically four coexistence generations at this time (excluding the older 4700/5150and 5100M systems).

4800/5200 485x/525x 4900/5300 4950/5350 4980/5380

450 MHz 700 MHz 1.4 GHz 2.8 GHz 3.06 GHz 500 MHz 900 MHz 550 MHz

If one system has the 4800/5200 nodes with different speeds of CPUs (450/500/550), this isreferred to as co-residence, not coexistence. All of the CPUs within a single node must bethe same speed.

For example, a system with 4850/5250 nodes can co-reside with 4851/5251 SMPs. Asystem with 4850/5250 nodes can co-exist with 4900/5300 SMPs.

General notes:- There is no BYNET V1.1 coexistence with 485x/525x - a 4800 system will have to

be upgraded to BYNET V2.1 in order to coexist with a 485x.

- The general purpose of the 4851/5251 systems is to co-reside with the 4850/5250systems.

- 4800/5200 SMPs can be upgraded to 4850/5250 SMPs. This is primarily a CPUupgrade

- There is no field upgrade from 700 MHz CPUs to 900 MHz CPUs for 4850/5250SMPs.

- 4900/5300 SMPs can be upgraded to 4950/5350 or 4980/5380 SMPs. This is achassis upgrade (or swap).


NCR Coexistence Combinations

4800/5200Pentium Xeon 450/500/550 MHz

4900/5300Pentium III Tualatin 1.4 GHz

Co-residence =this family of

products

Co-residence =this family of

products

NCR provides investment protection to customers by allowingcoexistence of different generations in the same system.

These families ofproducts can co-exist in one system

These families ofproducts can co-exist in one system

4850/5250Pentium Xeon 700 MHz



4950/5350Pentium IV Xeon 2.8 GHz

4980/5380Pentium IV Xeon 3.06 GHz

Sometimes referredto as a

“performance class”



What is the BYNET™?The BYNET (BanYan Network) provides high performance networking capabilities forNCR MPP systems. The BYNET is a dual-redundant, bi-directional, multi-staged networkbased on a Banyan network topology. The BYNET enables multiple processing nodes(SMP nodes) to communicate in a high speed, loosely-coupled fashion.

BYNET communication occurs in a point-to-point, multi-cast, or broadcast fashion. Aconnection request contains an address or routing tag for the intended receiving node orgroup of nodes. Once the connection is made, a circuit is established for the duration of theconnection. The BYNET works much like a telephone network where many callers canestablish connections, including conference calls.

The BYNET Version 1 interconnect provides a peak bandwidth of 10 Megabytes (MB) persecond for each node connected to a network. The overall network bandwidth scaleslinearly as nodes are added to a system network.

The BYNET Version 2 interconnect provides a peak bandwidth of 60 Megabytes (MB) persecond for each node connected to a network.

For example, an 8 node 5350 system with a dual BYNET network has the potential rawcapability of 960 MB per second total bandwidth for point–to–point connection. Totalavailable broadcast bandwidth is 120 MB per second for a dual network system of any size.

Other features of the BYNET network include:

• Guaranteed delivery - a message from a node is guaranteed to be delivered withouterror to the receiving node(s); multiple levels of error checking andacknowledgment are used to ensure this.

• Fault tolerant - multiple connection paths are available in each network; dualnetwork feature provides an active backup network should one network be lost.

• Flexible network usage - nodes communicate in point-to-point or broadcast fashion.

• Self-configuring - the BYNET automatically determines network topology at start-up; enables ease of installation.

• Self-diagnosis and automatic fault recovery - automatically detects and reportserrors; reconfigures routing of connections to avoid inoperable processing nodes.

• Load balancing - traffic is automatically and dynamically distributed throughoutthe networks.

NOTE: In this course, the designation BYNET V1 groups the features of BYNET Versions1.0 and Version 1.1 into one category (BYNET V1). BYNET V2 groups the features ofBYNET V2.0 and V2.1 together.


What is the BYNET?

BIC (BYNET Interface Card)

• BIC2M - has 2 BYNET v2.1ports; e.g., used with 4980

• BIC4M - has 4 BYNET v2.1ports; e.g., used with 5380

BYNET Switches• BYNET 4 switch (Version

2.1, 60 MB/sec)

• BYNET 16 or 64 switches(Version 2.0, 60 MB/sec)

. . . .

BYNET V1 or V2 Switch

BYNET V1 or V2Switch

BIC

SMP

Open BYNET SW

SMP

Open BYNET SW

BIC

What is the BYNET (BanYan NETwork)?• High speed interconnect (network) for processing nodes in NCR MPP

systems. The BYNET is a dual redundant network.• BYNET works much like a telephone network where many callers (nodes) can

establish connections, including conference calls.



BYNET 4 and BYNET 16 SwitchesThe facing page contains of examples of BYA4M and BYA16G switches. Details on theseand other BYNET switches are listed below.

BYNET 4 Switch (BYA4P) - a PCI card designed to interconnect up to 4 SMPs. Thisswitch is effectively a BYNET V1.1 switch (10 MB/sec.) and is used in the 4800. TheBYA4P is a PCI card that is placed into a PCI slot of an SMP.

BYNET V2 4 Switch (BYA4G) - PCI card designed to interconnect up to 4 SMPs. Thisswitch is a BYNET V2 switch (60 MB/sec.) designed for 485x systems. The BYA4G is aPCI card that is placed into a PCI slot of an SMP.

BYNET V2.1 4 Switch (BYA4M) - PCI card designed to interconnect up to 4 SMPs. Thisswitch is a new BYNET V2.1 switch (60 MB/sec.) designed for 4900 systems. TheBYA4M is a PCI card that is placed into a PCI slot of an SMP.

BYNET V2 16 Node Switch (BYA16G) – this V2 switch (60 MB/sec.) allows up to 165200 SMPs to interconnect. This 3U chassis switch resides in the 5200/5300 SystemCabinet. With the 5380 release, a 1U BYA16G switch is available.

4700/5150 BYNET Switches (not shown)NCR 4700 and 5150 systems used BYNET Version 1.1 switches. These switches are notshown in this module, but are briefly described below.

There are three types of BYNET V1.1 switches. The main difference between the BYNET8, the BYNET 16, and BYNET 128 chassis is the switch board in the chassis. The rear ofthe BYNET chassis has different types (and number) of connections depending on the typeof switch board.

BYNET 8 (BYA8) Chassis - designed to allow up to 8 processing nodes to interconnect.This will be used for systems that do not have the need for scalability beyond 8 nodes. Thisswitch is used with 4700 systems.

BYNET 16 (BYA16) Chassis - designed to allow up to 16 processing nodes to interconnect.This switch is used with 5150 systems.

BYNET 128 (BYB32) Chassis - designed to allow up to 128 processing nodes tointerconnect. The BYB32 switch connects multiple BYA16 switches together. For every 2BYA16 switches, you need 1 BYB32 switch. This switch is used with 5150 systems.


BYNET 4 and BYNET 16 Switches

SMP4

SMP3

SMP2

SMP1

BYNET 0

BYA4M

BYNET 1

BYA4M

BYNET 0 BYNET 1

. . .SMP2

BYA16G Switch BYA16G Switch

SMP3

SMP1

SMP4

SMP16

• A BYNET 4Switch is a PCI-based cardlocated insidean SMP.

• A BYNET 16switch is aseparate 3Uchassis locatedinside the SMPcabinet.



BYNET 64 SwitchesFor configurations greater that 16 nodes, the BYNET 64 switch must be used.

BYNET V2 64 Node Switch (BYA64GX chassis) – this V2 switch is actually composed of8 BYA8X switch boards in the BYA64GX chassis. Each BYA8X switch board allows upto 8 SMPs to interconnect (i.e., 8 switches x 8 SMPs each = 64 SMPs). The BYA64GX isactually a backpanel that allows the 8 BYA8X switch boards to interconnect. TheBYA64GX also includes a Diagnostic Processor (DP) board. This 12U chassis resides ineither the BYNET V2 64 Node Switch cabinet or the BYNET V2 64/512 Node ExpansionCabinet.

Note: BYA8X switch board (in BYA64GX chassis): This is Stage A base switch board.Each board supports 8 links to nodes. The BYA64GX chassis can contain amaximum of 8 BYA8X switches, allowing for 64 links to nodes. In systems greaterthan 64 nodes, the BYA8GX switch boards also connect the BYA64GX chassis toBYB64G chassis through X-port connectors, one on each BYA8X board.

BYNET V2 Switch CabinetsThe BYNET V2 64 Node Switch Cabinet (shown on facing page) can be used forconfigurations from 2 through 64 nodes and must be used for configurations greater than 16nodes. All nodes in the configuration are interconnected from the BYNET V2 nodeinterface to the BYNET V2 64 Node Switch chassis (BYA64GX). Two BYNET V2 64Node Switch Cabinets are required for the base dual redundant BYNET V2 networks.

The BYNET V2 512 Node Expansion Cabinet (not shown) is for used for configurationsthat begin with 64 nodes or less and has expanded beyond 64 node maximum configurationsupported by the BYNET BYA64GX chassis (in the V2 64 Node Switch Cabinet). Above64 nodes, the BYNET V2 BYB64G chassis (effectively a 512 node switch chassis) is usedto interconnect multiple BYNET V2 64 node switch chassis. The simple configurationrules are:

• Each group of 2 to 64 nodes requires two BYNET V2 64 node switch chassis; aminimum of two is required for dual redundancy.

• For configurations with greater than 64 nodes, each BYNET V2 64 node switchchassis must have a complimentary BYNET V2 512 node switch chassis.


BYNET 64 Switches

• A BYNET 64 Switch is a separate 12Uchassis located inside a BYNET rackor cabinet.

• Two BYNET switch racks are neededto house these two BYNET 64switches.

SMP1

SMP2

SMP64

. . .

BYA Switch(BYA64GX)

BYNET 0 BYNET 1

SMPs connect toBYA switches.

BYA Switch(BYA64GX)

BYNET Switch Cabinet or Rack

BYNET V264 NodeSwitch

Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch



BYNET Expansion Switches (BYB64G)The facing page illustrates the purpose of BYA64GX and BYB64G switches. BYAswitches connect SMPs. With more than 64 SMPs, multiple BYA switches are needed forone BYNET and the multiple BYA switches are connected together with BYB switches.

The BYB switches are called BYNET Expansion switches.

BYNET Expansion Switch (BYB64G chassis) – this V2 switch is actually composed of 4BYB16G switch boards in the BYB64X chassis. This is effectively the Stage B expansionswitch board. These boards interconnect through the BYB backpanel to provide 8expansion ports. The expansion ports are used to interconnect BYA64GX chassis (throughX-port connections on BYA8X switch boards). A maximum of 8 BYA64GX chassis can beinterconnected to a maximum of 8 BYB64G chassis (one X-port connection from eachBYA64 chassis to the expansion port on each BYB64 chassis), thus providing up to 512-node switching capacity. This chassis also include a Diagnostic Processor board.


BYNET Expansion Switches (BYB64G)

• A BYNET Expansionswitch (BYB) is aseparate 12U chassislocated inside theBYNET rack orcabinet.

• BYNET ExpansionSwitches connectBYNET (BYA) 64switches together toallow for expansionof a system up to512 nodes.

SMP128

SMP1

SMP2

SMP64

SMP65

SMP66

. . .. . .

BYNET 1

BYA Switch(BYA64GX)

BYA Switch(BYA64GX)

BYB Switch(BYB64G)

BYB Switch(BYB64G)

BYNET 1 is not shown, but all SMPs connect to both BYNETs.



BYNET Expansion Switches (cont.)The facing page illustrates a system with 128 nodes and the BYNET switches that would beused. BYNET 64 and BYNET Expansion switches are placed in a separate BYNET cabinet.

The BYNET V2 64/512 Node Expansion Cabinet is used for configurations that beginwith greater than 64 nodes. Two of the cabinets provide the BYNET V2 switchingcapability of two BYNET V2 64 Node Switch Cabinet and two BYNET V2 512 NodeExpansion Cabinets.

The SLAN and PvtLAN switches are located in the back of the cabinet and do not appear aschassis modules on the AWS. These networking components are used to connect the AWSto the system.

The SLAN switch is an 8 port 10BaseT switch and is used to connect the BYNET diagnosticprocessors to the SLAN (System LAN).

The PvtLAN switch is a 24 port 10/100 BaseT switch and is used to connect multiplePvtLAN hubs together for the PvtLAN (Private LAN). The PvtLAN switch is a 2Ucomponent.


BYNET Expansion Switches (cont.)

SMP2

SMP64

SMP128

SMP66. . .

BYNET 1

. . .SMP1

SMP65

BYB Switch(BYB64G)

BYA Switch(BYA64GX)

BYA Switch(BYA64GX)

BYB Switch(BYB64G)

BYB Switch(BYB64G)

BYA Switch(BYA64GX)

BYA Switch(BYA64GX)

BYB Switch(BYB64G)

BYNET 0

BYNET Switch Cabinet or Rack


Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

BYNET V2Expansion

SwitchChassis

(BYB64G)

This example shows both BYNETs.



BYNET™ SoftwareThe bynet software package is required for the Teradata Database. This package containsthe blm (BYNET Link Manager) and bdl (BYNET Data Link) drivers. This packageprovides for both BYNET protocol and TCP/IP communication across the BYNET.

BYNET™ Device DriversThe bynet software package provides two primary drivers to access the BYNET.

• blm – BYNET Link Manager• bdl – BYNET Data Link

BLMThe BLM software driver is a STREAMS-based UNIX driver and is responsible formanaging the BIC adapters (physical device). The BLM driver places messages/data on theBIC adapters and is the UNIX interface to the BYNET. This driver is linked into the UNIXkernel.

The following software currently uses the BLM driver:

• Teradata database software to directly access the BYNET.

• BDL software driver – interface between TCP/IP software and BLM.

• Can also be used by databases or applications other than Teradata.

There are two blmd daemons running on an SMP node – one for each BYNET. Thefunctions of these daemons include:

• Topology generation when UNIX initializes and when topology needsregenerating.

• Handle heartbeat messages.

BDLThe BDL software driver is also a STREAMS-based UNIX driver and its primary purpose isto allow TCP/IP over the BYNET.


BYNET™ Software

BYNET software package (bynet)

• provides the blm and bdl drivers forthe BYNET.

– Provides features required byTeradata.

• Simple (UNIX) diagnostic utility is

# /etc/bin/bam

# bam -sBoard State Nodes Net # Net nameBPCI 0 Online 2 0 BYA022-0BPCI 1 Online 2 1 BYA022-1

Teradata User TCP/IP User

BDL

BLM

TCP UDP

IP

PCI bus

Teradata DBMS

BYNET 0 BYNET 1

BYNET I/F 2

BYNET I/F 3BIC4M

BYNET I/F 1

BYNET I/F 0



Administration Workstation (AWS)The AWS provides a single operational view of an NCR Large System (3600, 4700, 4800,4850, 4855, 4900, 4950, 5100, 5150, 5200, 5250, 5255, 5300, and 5350), and theenvironment to configure, monitor, and manage the system. The AWS is a UNIX-based orWindows-based processor with a user-friendly graphical interface.

The AWS effectively becomes a central console for NCR MPP systems.

• 3600 - multiple nodes to support UNIX and the Teradata database.

• 5100C - Clustered environment of multiple UNIX nodes

• 5100M - MPP (Massively Parallel Processing) environment

• 4700/5150/48xx/52xx - Rack mount, multiple UNIX node system

• 4900/5300 – Rack mount; up to 4 SMPs per rack or cabinet



Additional AWS Capabilities

Dual or Multiple AWSA redundant AWS, referred to as “Dual AWS”, is an available option. It provides aredundant system console to the system. Both AWS consoles are totally independent, eachhaving full stand-alone functionality.

AWS software allows both AWS consoles to operate on the same SLAN simultaneously forall unrestricted functions. Security lockout software only allows one AWS at a time toperform secured functions (e.g., Power on/off).

Remote AWSThe AWS display can be extended for multiple administrators and operators. Remoteworkstations or PCs with this remote capability are sometimes called AdministrationStations (AS).


Administration Workstation (AWS)

The AWS is a standalone UNIX or Windowsbased workstation that is the primaryoperations interface (console) for WorldMarkMPP systems.

The AWS provides a single, graphical view ofthe system.

AWS

consolemessages

ETHETH

Cabinet 1

SMCBYNET 16

SMP

SMP

SMP

SMP

Cabinet 101

SMC

Disk Array

Disk Array

Cabinet 2

SMCBYNET 16

SMP

SMP

SMP

SMP

Cabinet 102

SMC

Disk Array

Disk Array

AWS provides GUI to ...• connect to SMPs• connect to Teradata DB

Window• power on/off/reset

components• manage faults• obtains h/w or s/w

status information



AWS Connectivity to a SystemTo communicate with processing nodes and system components, the AWS is connected tothe system via two Ethernet LANs.

• SLAN - System Ethernet LAN known as the SLAN; connects AWS to the CMIC(Chassis Management Interface Controller) in the Server Management Chassis ofeach cabinet

• PvtLAN – Private Ethernet LAN - connects AWS directly to SMPs.

System LAN (SLAN)The System LAN (SLAN) is a private Ethernet LAN between the AWS and CMICs in a48xx/52xx and 4900/5300 systems. The CMIC (Chassis Management Interface Controller)is the intelligent component (actually an Intel processor) of the SMC. The CMIC providesnetwork connectivity for the SMC to the SLAN.

The MLAN (Management LAN) is an internal LAN that connects components to the CMICsand also connects the SMC in a storage rack to the SMC in a processor rack.

• System Events• Console and Diagnostic connections

Private LAN (PvtLAN)This Ethernet LAN is used to directly connect the AWS to each processing node (SMP).

• All SMPs are connected using 10baseT (Twisted Pair).

• Up to 100 Mbps full duplex to AWS.

• The PvtLAN uses hubs and switches to allow connection to all nodes. SMP nodesare connected using 8 port 10BaseT hubs (Allied Telesyn MR820TR). A 24 port10/100BaseT switch (Allied Telesyn 8124XL) is used from the AWS to each 8 porthub.


AWS Connectivity to a System

AWS

consolemessages

ETHETH

Cabinet 1

SMCBYNET 16

SMP

SMP

SMP

SMP

Cabinet 101

SMC

Disk Array

Disk Array

Cabinet 2

SMCBYNET 16

SMP

SMP

SMP

SMP

Cabinet 102

SMC

Disk Array

Disk Array

SLAN

PvtLAN

MLAN MLAN

AWS is connected to the system via two Ethernet LANs.• SLAN (System Ethernet LAN) – connects AWS to the CMIC in the SMC of each

processor cabinet.– Used for hardware functions; console connections, power control, etc.

• PvtLAN – Private Ethernet LAN - connects AWS directly to SMPs.– Used for software functions; LAN connections, Teradata Database Window, etc.



AWS Naming ConventionsThe examples on the facing page show AWS naming conventions for NCR cabinets orracks. Each chassis consists of a number of internal components (processors, fans, powersupplies, management boards, etc.).

A 4950/5350 system can have up to 10 chassis modules in a cabinet.

Within each chassis, the common naming conventions for the chassis components are asfollows:

• Power Supplies - PS1, PS2, PS3

• Fans - FAN 1, FAN 2, FAN 3, FAN 4, FAN 5, FAN 6, FAN 7, FAN 8

• Components within the SMC

CMIC (Chassis Management Interface Controller)CMB Rack for SMC (Server Management Chassis) itselfPTB (Pass Through Board) for BYNET and SMP chassisesRCMB (Remote CMB) for each UPS/Power SupplyUMB (Universal Management Board) for each SMP chassisCMB (Chassis Management Board) within each BYNET chassis

On AWS processor level display, each of the “management boards” is identified as a MB(Management Board).

• Power Supplies and Batteries

UPS 1, UPS 2, UPS 3


AWS Naming Conventions

5250 Rack

CMIC001-1

PWR001-6

PWR001-7

PWR001-8

SMP001-4

SMP001-5

BYAS001-2

BYAS001-3

5380 Rack

CMIC002-1

BYAS002-2

BYAS002-3

PWR002-8PWR002-9PWR002-10

SMP002-4

SMP002-5

SMP002-6

SMP002-7

BYNET Rack

CMIC003-1

BYAS003-2


Chassis(BYA64GX)

BYB003-3

BYNET V2Expansion

SwitchChassis

(BYB64G)

PWR003-9PWR003-10



AWS System Display (UNIX)The facing page contains an example of an AWS system display of a 4950. SMP cabineticons for a 4900/5300 and 4950/5350 cabinet can show up to 4 SMPs in the cabinet. Thisdisplay consists of two windows:

• Administration Workstation window

• CSF (Customer Support Facility) Fault window

AWS System WindowThe system screen provides the following features and functions:

• Graphical representation of the system

• Status of system components

• Interface from which you can start all administrative applications (schedule tasks,perform backup, subsystem maintenance, power control, etc.)

• Diagnostics and replacement of Field Replaceable Unit (FRU) components

CSF Faults WindowThe CSF Faults Window provides access to outstanding faults and messages in the system.This window is also used to customize CSF and use the monitoring and tracking features ofCSF.

The Fault window is a small, resizable and movable window that is automatically displayedon the AWS. When you click the Faults button of the Fault window, the Fault List windowwill open and show the currently outstanding faults with one line of information.

From this window, you can select menu commands to perform such activities as viewingfaults, deleting faults, reporting problems and reviewing recommended solutions.

The Fault List window has a filtering function that allows you to filter out specified faults.For example, if the filtering option is used, the Fault window may contain a value of 12 of16 indicating that there are 12 faults appearing in the Fault List window, but there are really16 faults.


AWS System Display (UNIX)

NCREnterpriseStorageRack

SMP Rack



AWS Processor Level Display (UNIX)Two levels of display are provided in the Administration Workstation window, cabinet leveland processor level. The level is identified in the upper left area of the window.

• The CABINET Level display displays one icon for each cabinet (previous page).

• The PROCESSOR Level display shows one icon for each processor within theselected cabinet (e.g., facing page). This level is applicable to all cabinet iconsexcept the AWS.

By right clicking on a cabinet icon, the AWS display will change to a “Processor LevelView”. In the example on the facing page, an icon represents each chassis in a 4950cabinet.

Each chassis also has an associated “management board” which is represented with the MBicon.

To change from the PROCESSOR Level to the CABINET Level display, either click orright-click within the shaded area of the UP LEVEL icon.


AWS Processor Level Display (UNIX)



Windows 2000 AWS ExampleThe facing page contains an example of a Windows 2000 AWS display.


Windows 2000 AWS Example


Summary

• 48xx, 49xx, 52xx, and 53xx systems all utilize an rack-based cabinetarchitecture.

• SMPs and components are housed in chassis modules and are mountedin an industry standard rack.

• The 4950/5350 and 4980/5380 systems use the latest Intel architecturesand CPUs (Pentium IV Xeon) to provide the fastest processing nodes inthe NCR systems family.

• The BYNET™ is the high speed interconnect (network) for SMP nodes inNCR’s WorldMark MPP systems.

– BYNET V2 and V2.1 - 60 Megabytes (MB) per second per BYNETnetwork.

• The AWS is the primary operations interface (console) for WorldMark MPPsystems.


Review Questions

Answer the following questions:

1. Specify 2 improvements of the 53xx systems as compared to the 52xx systems. ________________________________________________________________________________________________________________________________

2. What does the acronym represent and briefly define the purpose of the following subsystems?

SMP ________________________________________________________________________________

SMC ________________________________________________________________________________

BYNET ________________________________________________________________________________

AWS ________________________________________________________________________________

3. List the two LANs that connect the AWS to a 49xx/53xx system and briefly describe the purpose ofeach LAN.

________________________________________________________________________________________________________________________________________________________________

4. Specify the names of the SMPs in 5380 SMP cabinet #3._______________ _______________ _______________ _______________



Notes

NCR 49xx/53xx Configurations and Disk Arrays Page 12-1

Module 12


NCR 49xx/53xx Configurations and Disk Arrays


• Specify 3 improvements of the 5380 SMP as compared to the5350 SMP.

• List an improvement of the 6841-2456 as compared toprevious 6840-1456 disk array.

• List an improvement of the 6841-6456 as compared to 6841-2456 disk array.

• Identify sizes of components and configurations.


Page 12-2 NCR 49xx/53xx Configurations and Disk Arrays

Notes


Table of Contents

NCR 4950/4980 AND 5350/5380 SYSTEMS........................................................................................................ 12-4SYSTEM FEATURES AND ENHANCEMENTS .............................................................................................. 12-6

4950/4980 SYSTEMS............................................................................................................................................ 12-65350/5380 SYSTEMS............................................................................................................................................ 12-6

4950/5350 AND 4980/5380 PROCESSING NODE (OR SMP) .......................................................................... 12-84950/5350 AND 4980/5380 MEMORY.................................................................................................................. 12-8

4950/5350 AND 4980/5380 INTEGRATED PERIPHERALS.......................................................................... 12-104950/5350 AND 4980/5380 PCI I/O SUBSYSTEM ........................................................................................... 12-12BASE AND EXPANSION RACKS .................................................................................................................... 12-14EXAMPLE 1: NCR 4980 SYSTEM ................................................................................................................... 12-16EXAMPLE 2: 5380 SYSTEM – 8 NODES ........................................................................................................ 12-18EXAMPLE 3: 5380 SYSTEM – 16 NODES ...................................................................................................... 12-20EXAMPLE 4: 5380 SYSTEM – 32 NODES ...................................................................................................... 12-22DISK ARRAY PROVIDERS.............................................................................................................................. 12-24NCR ENTERPRISE STORAGE SOLUTIONS................................................................................................ 12-26NCR ENTERPRISE STORAGE CABINETS................................................................................................... 12-28NCR 6288-1440 SCSI DISK ARRAY................................................................................................................. 12-30NCR 6288-1452 HIGH DENSITY DISK ARRAY............................................................................................ 12-30NCR 6841-2456 FIBRE CHANNEL DISK ARRAY ........................................................................................ 12-32NCR 6841-6456 FIBRE CHANNEL DISK ARRAY ........................................................................................ 12-345380 AND 6841-2456 DISK ARRAYS ............................................................................................................... 12-365380 AND 6841-6456 DISK ARRAYS ............................................................................................................... 12-38PERFORMANCE VERSUS CAPACITY ......................................................................................................... 12-40SUMMARY .......................................................................................................................................................... 12-42REVIEW QUESTIONS....................................................................................................................................... 12-44



NCR 4950/4980 and 5350/5380 SystemsNCR MPP systems represent NCR’s large system configurations to support the TeradataDatabase. The NCR 4980 and 5380 systems – projected release date of April 30, 2003 -represent NCR’s newest systems to support the Teradata database. The 4950/5350 systemswere released on December 20, 2002.

The NCR 4950/4980 systems can scale up to 4 SMPs. The NCR 5350/5380 systems canscale up to 512 SMPs.

NCR 4950/4980 systems use a BYNET 4 switch that allows a maximum of 4 SMPs. TheBYNET4 switch uses is actually a PCI card that fits into a PCI slot of one of the SMPs.

The NCR 5350/5380 systems can scale all the way up to 512 nodes depending on theBYNET V2 switches used.

For configurations up to 16 SMP nodes, BYNET V2 16 chassis switches (BYA16G) may beused. These switches are self-contained in a 1U (for 5380) or 3U chassis and reside in the5350 System cabinet. The 1U BYNET 16 switch chassis is new with the release of the5380.

For configurations above 16 SMP nodes, the BYNET Switch Cabinets that contain BYNET64 switches are required.

Each 4950/5350 SMP utilizes two Intel Pentium IV Xeon 2.8 GHz CPUs and a 400 MHzFSB (Front Side Bus).

Each 4980/5380 SMP utilizes two Intel Pentium IV Xeon 3.06 GHz CPUs and a 533 MHzFSB (Front Side Bus).

Characteristics or both systems include:

• Shared memory – standard 2 GB; can be configured with up to 4 GB for Teradatanodes (TPA nodes).

• I/O architecture – PCI, 3 PCI buses• 4 internal 18 GB disks and removable media devices


NCR 4950/5350 and 5350/5380 Systems

• NCR 4950/5350 and 5350/5380 systems leverage a common hardwarebuilding block - the SMP node.

– Up to 4 SMPs can be placed in one rack or cabinet

– 4950/5350 SMPs utilize two Intel Pentium IV Xeon 2.8 GHz CPUs and a400 MHz FSB

– 4980/5380 SMPs utilize two Intel Pentium IV Xeon 3.06 GHz CPUs anda 533 FSB

• Multiple independent nodes (SMPs) work together as a system throughthese technologies.

– Teradata - parallel database– BYNET - system interconnect– AWS - single point of operational control

• NCR 4950 and 4980 systems scale up to 4 SMPs.

• NCR 5350 and 5380 systems scale up to 512 SMPs



System Features and EnhancementsThe facing page identifies features and enhancement for NCR 4950/4980 and 5350/5380systems. Some characteristics specific to each are listed below.

4950/4980 SystemsNCR 4950/4980 systems are designed for entry-level and mid-range data warehouseapplications (up to 4 TB of MaxPerm space). Characteristics include:

• Consists of a maximum of 4 processing nodes interconnected with the BYNET.

• Utilizes redundant BYNET 4 PCI switches to support up to 4 processing nodes.BYNET 4 switches are located within an SMP – lower cost entry point than otherBYNET switches.

• BYA4M switch – Version 2.1 (60 MB/sec. per BYNET)

• 4950/4980 systems can be upgraded to 5350/5380 systems.

– Requires different BYNET switches

• Existing 4900 SMPs (in 4900 racks) can be upgraded to 4950/4980 SMPs.

– SMP chassis replacement

5350/5380 SystemsNCR 5350/5380 systems are designed for data warehouse applications accessing data from 1TB to greater than 100 TB. Characteristics include:

• Consists of a maximum of 512 processing nodes interconnected with the BYNET.

• Utilize redundant BYNET switches.

– BYNET 16 switches - support up to 16 processing nodes.

– BYNET 64 switches - support up to 64 processing nodes.

– BYNET 64 and BYNET Expansion switches - support up to 512 processingnodes.


System Features and Enhancements

System features:

• Up to 4 SMPs or nodes can reside in a rack - more power in the samefootprint with Intel Pentium IV Xeon 2.8 or 3.06 GHz CPUs– Improved system MTBF - four nodes per cabinet usually means fewer cabinets.– Expandable In-cabinet upgrade; add nodes to partially populated rack

• TPA nodes running MP-RAS can be integrated with non-TPA nodesrunning Windows 2000 in same system.

• Uses existing BYNET 2.0 fabric switches; uses BYNET BIC Release 2.1interface cards - increases the PCI I/O throughput of the BICs.– With the release of the NCR 5380, a new BYNET 16 (BYA16G) switch 1U switch

chassis is also released and utilized.

• Server Management features– utilizes CMIC3 (4980/5380 utilize a new 11 slot SMC (vs. previous 10 slot SMC)– uses 2U UPS chassis with integrated IS (Input Selector)

• AWS capabilities– Windows 2000 AWS for MP-RAS systems– AWS can monitor mixed operating systems in a single MPP system



4950/5350 and 4980/5380 Processing Node (or SMP)The Processing node is comprised completely of Intel designed components. The 4950/5350components consist of an Intel Hudson III chassis kit and an Intel Hodges boxed board setkit. The 4980/5380 systems utilize the Intel Harlingen boxed board set kit. NCR replacesthe existing color front panel with an NCR panel.

A description of key components is:

• Intel Hudson III Node Chassis – chassis used for node

• Intel Hodges (4950/5350) Base Board – SHG2– 2-Way Pentium 4 Xeon processor 2.8 GHz– Integrated ATI PCI video– Integrated Adaptec dual channel Ultra 160 SCSI– Integrated dual 10/100Mb Ethernet– Six 72-bit sockets for 184-pin, 200 MHz, 2.5 V, DDR200 or DDR266

compliant, registered, ECC, SDRAM single-sided or double-sided memorymodules (DIMM)

– Entry server SSI/ATX form factor (12” x 13”)– 6 full length PCI slots

• The 4980/5380 is very similar except that it utilizes the Intel Harlingen boxedboard set (SE7501HG2) and uses two 3.06 GHz CPUs and has a 533 MHz FSB.

• Note: Some of the newer 4950/5350 processing nodes may also utilize theHarlingen baseboard, but these nodes will have the FSB bus speed set to 400 MHz.

4950/5350 and 4980/5380 MemoryThe baseboard provides six DIMM sockets supporting three pairs of DIMMs: PC1600(DDR200) for the 4950/5350, upward compatible with PC2100 (DRR266) for the4980/5380. Memory is partitioned in three banks and can be implemented with eithersingle-sided (one row) or double-sided (two rows) DIMMs, allowing for a maximummemory capacity of 12 GB using 2-GB DIMMs.

NCR 4950/535 and 5350/5380 nodes support the following maximum memoryconfigurations:

• 4 GB for all MP-RAS nodes (TPA and non-TPA)

• 4 GB for Windows TPA nodes

• 6 GB for Windows non-TPA nodes


4950/5350 and 4980/5380Processing Node (or SMP)

• Basic building block for 4950/5350 and 4980/5380 systems. Other names -compute node, 2-way node, or SMP.

• 4950/5350 Processing node characteristics:– 2 Intel Pentium IV Xeon CPUs - 2.8 GHz– 400 MHz internal FSB (Front Side Bus)

• 4980/5380 Processing node characteristics:– 2 Intel Pentium IV Xeon CPUs - 3.06 GHz– 533 MHz internal FSB (Front Side Bus)

• Processing node characteristics of both nodes:– 5U (height) chassis - allows 4 nodes within a rack– 2 GB memory, optionally 1 - 4 GB memory for Teradata– 3 PCI buses - fast (133 MHz) and wide (64-bit)– Integrated disk and local media devices– 1 BIC PCI interface adapter– Integrated dual 10/100 Ethernet adapters

• Can be added and/or upgraded in the fieldU = unit of measurement – 1.75"



4950/5350 and 4980/5380 Integrated PeripheralsThe node has a hot-swap SCSI drive bay that supports five SCSI hard disk drives with thefollowing specifications:

• 3.5-inch• Single connector attach (SCA2)• Low voltage differential (LVD)

The drives are mounted on drive carriers for hot-swap insertion. Unpopulated slots containcarriers with air baffles.

The SCSI drives are numbered 0– 4 from bottom to top in the SCSI drive bay chassis, asshown on the facing page. LEDs on each drive indicate activity and fault.

Details on the front panel are:

Feature FunctionA Hard drive activity LEDB Power buttonC Sleep buttonD LAN #1 activity LEDE Reset buttonF ID toggle switchG ID LEDH LAN #2 activity LEDI Status LEDJ Power/Sleep LEDK NMI (non-maskable interrupt) switch

Note: Presented on the faceplate through a tiny hole. A small pointed object (like a paperclip) is needed to engage the switch.


4950/5350 and 4980/5380 Integrated Peripherals

3.5” Diskette Drive

CD-ROM

8mm Tape

Optional Disk

System Disk

System Disk

System Disk

Boot Disk

4

3

2

1

0

SMP has 4 (standard) 18 GB or 36 GB integrated disks that hold …

• Operating System (e.g., UNIX MP-RAS or Windows 2000)

• Teradata RDBMS software

Optional 5th disk can be 18 GB or 36 GB.



4950/5350 and 4980/5380 PCI I/O SubsystemPIC Adapters available for the 4950/49805350/5380 processing node include:

External fibre channel disk or tape storage

– LSI Logic quad fibre channel 2 Gbit (LSI7004G2)– QLogic quad fibre channel 1 Gbit (QLA2204F)

External SCSI and high-density (HD) SCSI disk storage (UNIX MP-RAS andWindows):

– NCR high performance PQS (HP-PQS 80 MB/s)– NCR PCI Quad SCSI (HP-PQS 40 MB/s)

External SCSI tape storage:

– NCR PCI Quad SCSI (HP-PQS 40 MB/s)

Communications:

– Gigabit Ethernet optical or copper– Dual Ethernet 10/100BaseT (PCI)– Token Ring PCI– Fiber Distributed Data Interface (FDDI) PCI, dual-fiber

Mainframe interface:

– PCI Bus-to-ESCON Adapter (PBSA): serial channel interface (EnterpriseSystems Connection – ESCON)

BYNET interconnect: BIC2M (4950/4980), BIC4M (5350/5380)


4950/5350 and 4980/5380 PCI I/O Subsystem

The baseboard provides six PCI expansion slots in the node chassis, asfollows:

• Two 64-bit/100 MHz slots• Three 32-bit/33MHz slots• One 64-bit/133 MHz slot

Examples of PCI Adapters:• LSI Quad FC (Fibre Channel) host adapter - provides 4 Fibre Channel

connections (2 Gbit/sec.) - connects SMPs to disk arrays (e.g., 6841-2456).

• High Performance PQS host adapter - provides 4 Ultra/Wide 2 SCSIbuses (80 MB/sec.) - connects SMPs to SCSI based arrays

• BIC2M for 4950/4980; connects to the BYA4M switches• BIC4M for 5350/5380; connects to BYNET v2.0 switches• PCI Bus ESCON Adapter (PBSA) - connects an ESCON channel.• Networking adapters - e.g., Ethernet 10/100 BaseT, GigaBit Ethernet



Base and Expansion RacksNCR 4950/5350 and 4980/5380 systems use industry standard rack mount architecture andindividual chassis that conform to industry standards. Types of chassis that can be placed ina 49505350 or 4980/5380 cabinet include:

• Processing Node• BYNET V2 16 Node Switch (BYA16G)• Server Management• Uninterruptable Power Supply

The rack is referred to as a 40U rack. A U represents a “unit of vertical measurement” in anindustry standard rack. 1U = 4.445 cm or 1.75” high. Therefore, this cabinet has 40U or 40x 1.75” (70”) of usable space. The following table lists the height of each chassis:

Chassis HeightService Management 3U (13.3 cm, 5.25 in)BYA16G 3U (13.3 cm, 5.25 in)

1U (4.445 cm, 1.75 in.)4950/5350 or 4980/5380 SMP Node 5U (22.2 cm, 8.75 in.)UPS – 4950/5350 or 4980/5380 2U (8.9 cm, 3.5 in.)

A Base (or System) rack contains either BYNET 4 or BYNET 16 switches and SMP nodes.An Expansion rack contains SMP nodes, but no BYNET switches.

Depending on the number of nodes required in the configuration:

• 1 Node: Use the 4950 or 4980 Single Node cabinet

• 2 – 4 Nodes: Use an appropriate combination of 4950/4980 series Single, Two,Three and Four Nodes cabinet configurations; one base cabinet is required toprovide the BYNET switches. A BYNET switch is always required in multi-nodesystems. The Internal BYNET switch (BYA4M) is used in two different nodes.

• 5 – 16 Nodes: Use an appropriate combination of 5350/5380 series Base andExpansion cabinets. One 5350 Base cabinet is required to provide the BYNETswitches.

• 17 – 64 Nodes: Use an appropriate combination of 5350/5380 series Expansioncabinets along with two BYNET 64 switch cabinets. Two BYNET 64 cabinets arerequired to provide redundant BYNET fabrics.

• 65 – 512 Nodes: Use an appropriate combination of 5350/5380 series Expansioncabinets along with an appropriate number of BYNET 512 switch cabinets. TwoBYNET 512 cabinets are required to provide redundant BYNET fabrics for every64 nodes. A 65 to 512 nodes system would require 16 cabinets for redundantBYNET fabric.


Base and Expansion Racks

NCR

SMP001-4BIC2M BYA4M

SMC

SMP001-5BIC2M BYA4M

SMP001-6BIC2M

SMP001-7BIC2M

NCR

SMP001-4BIC4M

SMC

SMP001-5BIC4M

SMP001-6BIC4M

SMP001-7BIC4M

NCR

SMP001-4BIC4M

SMC

SMP001-5BIC4M

SMP001-6BIC4M

SMP001-7BIC4M

BYA16G Switch

BYA16G Switch

4950/4980 Base Rack 5350/5380 Base Rack 5350/5380 Expansion Rack



Example 1: NCR 4980 SystemThe facing page contains an example of an NCR 4980 system. This example illustrates aconfiguration where 4 SMPs are Fibre-channel connected to 3 disk arrays.

This configuration from a Teradata perspective will be described in more detail later in thismodule.


Example 1: NCR 4980 System

Notes:• 4980 Cabinet with BYNET

4 switches.

• Each SMP has a BICadapter to connect to theBYNET 4 switches.

• 4980 systems can beupgraded to 5380systems.– Requires different

BYNET switches

• Existing 4900 SMPs (in4900 racks) can beupgraded to 4980 SMPs.– SMP chassis

replacement NCR Enterprise Storage 6841-2456(available August, 2003)

NCR

SMP001-4BIC2M BYA4M

SMC

SMP001-5BIC2M BYA4M

SMP001-6BIC2M

SMP001-7BIC2M

NCR

SMC - 2U

6841-2456

6841-2456

NCR

SMC - 2U

6841-2456



Example 2: 5380 System – 8 NodesThe facing page contains an example configuration of an 8-node 5380 system utilizing theNCR 6841-2456 disk arrays. With the NCR 6841-2456 disk array, a typical configuration isto configure cliques of 4 nodes sharing 3 disk arrays.

What additional racks (cabinets) are needed to upgrade this system to 16 nodes?

Answer:

2 additional 5380 “Expansion” racks, each with 4 SMPs3 additional NCR 6841-2456 Storage racks


Example 2: 5380 System – 8 Nodes

6841-2456 Disk Arrays

SMCBYNET 16BYNET 16

SMP

SMP

SMP

SMP

SMC

SMP

SMP

SMP

SMP

Clique 0

Clique 1

Clique 0SMP001-4SMP001-5SMP002-4SMP002-5


Cabinet 1 Cabinet 2

What racks are neededto grow this system to16 nodes?



Example 3: 5380 System – 16 NodesThe facing page contains an example configuration of a 16-node 5380 system utilizing theNCR 6841-6456 disk arrays. With the NCR 6841-6456 disk array, a typical configuration isto utilize cliques of 4 nodes sharing 2 disk arrays.


Answer:

2 BYNET V2 64 Node Switch racks4 additional 5380 “Expansion” racks, each with 4 SMPs

4 additional NCR 6841-6456 Storage racks

The NCR 5380 system can scale up to 16 processing nodes using the BYNET V2 16 switch(BYA16G). This is a BYNET Release 2 implementation (i.e., 60 MB /sec).




SMCBYNET 16BYNET 16

SMP

SMP

SMP

SMP

SMC

SMP

SMP

SMP

SMP

Clique 0

Clique 1





Cabinet 1 Cabinet 2

What racks are needed to grow thissystem to 32 nodes?

SMC

SMP

SMP

SMP

SMP

SMC

SMP

SMP

SMP

SMP

Cabinet 3 Cabinet 4



Example 4: 5380 System – 32 NodesThe facing page contains an example configuration of a 32-node 5380 system utilizing theNCR 6841-6456 disk arrays. As mentioned earlier, with the NCR 6841-6456 disk array, atypical configuration is to configure cliques of 4 nodes sharing 2 disk arrays.

The NCR 5380 system can scale to 64 processing nodes using the BYNET V2 64 Nodeswitch (BYA64GX) which is housed in a BYNET V2 Switch cabinet. Note that there aretwo BYNET V2 Switch cabinets in the illustration on the facing page.

This is also a BYNET Release 2 implementation (i.e., 60 MB /sec).

What types of cabinets are needed to upgrade this system to 64 nodes?

Answer:

8 additional 5380 “Expansion” racks, each with 4 SMPs8 additional 6841-6456 Storage racks



SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

SMP

SMPSMP

SMP

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC

6841-6456

6841-6456

SMC


BYA64GX

SMC


BYA64GX

What additional racks are needed to upgrade this system to 64 nodes?




Disk Array ProvidersAs mentioned before, there are numerous disk array subsystems available to NCR largecomputer systems (e.g., 5380). Currently, disk array subsystems are primarily availablefrom two vendors – LSI Logic, Inc. and EMC2 .

Examples of disk array subsystem cabinets from LSI Logic are:

• NCR Enterprise Storage 6288 solution for UNIX MP-RAS systems – uses rack-based storage cabinet (similar to NSC) and supports 6288 disk array modules

– 6288-1440 – SCSI disk array supports up to 40 disks.

• NCR Enterprise Storage 6840 solution – uses rack-based storage cabinet andsupports 6840 disk array modules

– A 6840-1456 – Fibre Channel disk array supports up to 56 disks.

• NCR Enterprise Storage 6841 solution – uses rack-based storage cabinet andsupports 6841 disk array modules

– A 6841-1456 – Fibre Channel disk array supports up to 56 disks.

Examples of disk array subsystem cabinets from EMC2 are:

• 6276-2000 (replacement for 6273) – supports up to 32 disk drives (e.g., 18 GB 10Krpm); this disk array subsystem is intended for 4400 servers

• 6278-2000 (replacement for 6274) - supports up to 96 disk drives (e.g., 18 GB 10Krpm)

• 6278-3000 (Split backplane) - supports up to 96 (2 x48) disk drives (e.g., 18 GB10K rpm)

• 6290-8430-7300 - supports up to 96 18 GB disks• 6290-8431-7300 - supports up to 96 36 GB disks• 6290-8531 (replaces 6290-8431) – supports up to 96 36 GB drives• 6290-8231 (half-wide unit) – supports up to 48 36 GB drives


Disk Array Providers

Two primary storage vendors:• LSI Logic, Inc.• EMC2

Examples of NCR Enterprise Storage disk arrays from LSI Logic, Inc. are:

• Model 6288 SCSI - UNIX MP-RAS only– 6288-1440 supports up to 40 disks– uses a rack-based storage cabinet.

• Model 6841 Fibre Channel (UNIX MP-RAS)– 6841-2456 supports up to 56 disks– uses a rack-based storage cabinet

Examples of disk arrays from EMC2 are:

• 6290-8231 - supports up to 48 36 GB disks• 6290-8531 - supports up to 96 36 GB disks

Note: This module will provide examples of LSI Logic disk arrays.



NCR Enterprise Storage SolutionsThe chart on the facing page attempts to clarify the major NCR releases of LSI Disk Arrays.

Key for facing page:

NSC – NCR Storage CabinetWES – WorldMark Enterprise StorageDS – Deskside or PedestalMP-RAS – UNIX MP-RASWin NT – Windows NTWin 2000 – Windows 2000


NCR Enterprise Storage Solutions

LSI Disk Array Release Comparison

Release Key Feature Disk Array Class/Model O.S. Support GCA

NSC 1.0 Modular 6285-1220 MP-RAS 1998Arrays (SCSI) 6285-1440 Win NT, Win 2000

NSC 2.0 Fibre 6286-1220 Win NT 1999Channel Rack or DS Win 2000

WES 3.0 Quad Array 6288-1440 MP-RAS 1999(SCSI)

WES 3.5 Quad Array 6288-1440 MP-RAS 2000(SCSI) 6288-1452

WES 4.0 Fibre 6289-1440 Win 2000 07/2000Channel

WES 5.0 Fibre 6840-1440 MP-RAS 05/2002Channel 6840-1456

NCR Enterprise Fibre 6841-2456 (Performance Model) MP-RAS 08/2003Storage Channel 6841-6456 (Capacity Model) Win 2000 08/2003



NCR Enterprise Storage CabinetsThe NCR Enterprise Storage solution utilizes a modular array subsystem design. Forexample, the 6841-6456 will have a Controller Module and 4 Disk Drive Modules or DriveTrays.

One or two 6841-6456 disk arrays will be housed in an NCR Enterprise Storage cabinet.


NCR Enterprise Storage Cabinets

NCR Enterprise Storage Racks/Cabinets for Teradata MPP:

For TW 6.1

• 6288-1440/1452 Quad SCSI Arrays• 6289-1440 Quad Fibre Channel Arrays

For TW 6.2

• 6840-1440/1456 Quad Fibre Channel Arrays

For TW 7.1

• 6841-2456 Quad Fibre Channel Arrays• 6841-6456 Quad Fibre Channel Arrays

NCR

SMC - 2U

5884 Controllers

5884 Controllers

6841-6456



NCR 6288-1440 SCSI Disk ArrayThe NCR 6288-1440 (also known as WorldMark Enterprise Storage (WES)) solution uses arack-based cabinet. One or two 6288-1440 disk arrays are housed in the cabinet.

• Model 6288-1440 - consists of one 6288-4665 controller module and four 6288-1101 disk drive modules. The 1440 is 16U tall. This disk array can have up to 40disks.

This latest version (July, 2001) of this cabinet has an updated 2U SMC and three 2U UPSmodules.

NCR 6288-1452 High Density Disk ArrayThis disk array is similar to the 6288-1440, except each drive tray can hold 13 drives;effectively 13 x 4 = 52 disks in this array. This disk array is not shown on the facing page.

Characteristics include:

• New drive trays that support a total of 52 low profile hot swappable Ultra-2 SCSIdisk drives per array.

• Available disk drives: 18GB 10K rpm LP, 36GB 10K rpm LP, and 36GB 15K rpm(March, 2002).

Plus

• Dual hot swappable redundant RAID controllers with either 32MB or 256MB ofcache.

• SMPs use the High performance PCI Quad SCSI (HP-PQS) Adapter for Ultra-2SCSI point-to-point connectivity (80MB/sec) to multi-ported disk array.

• Support for RAID 1 and RAID 5.

And

• Fits into the same WES Cabinet as the 6288-1440.


NCR 6288-1440 SCSI Disk Array

One or two disk arrays can be placed inthe storage rack - a maximum of 80drives.This array is typically used with NCR48xx and 52xx systems.

Features include:• Each 4665 controller connects up to

4 SMPs using 4 point-to-point SCSI(80MB/sec.) buses.

• Dual AC Input and three 2U UPSunits (same as 4900/5300).

• Service Management - 2U SMCconnects to 3U SMC in SMP cabinet.

NCR

SMC - 2U

4665 Controllers

4665 Controllers

NCR

SMC - 2U

4665 Controllers



NCR 6841-2456 Fibre Channel Disk ArrayCabinet characteristics include:

• 2U Server Management Chassis

• Three 2U UPS with Dual AC Distribution Boxes

• Support for two arrays for a total of 112 drives and 4.1 TB raw data capacity using36 GB disk drives

Fibre Channel Disk Array characteristics include:

• Dual hot swappable redundant 4884 Fibre Channel RAID controllers with 1 GB ofcache and an Intel 550 MHz Celeron microprocessor.

• Each 6841-2456 disk array has support for up to 56 hot swappable 2 Gbit FibreChannel 36GB 15K rpm disk drives (Seagate Cheetah drives).

• Quad Fibre Channel Host Adapter (2Gb/sec) point-to-point connectivity betweenprocessing node and disk arrays.


Performance and availability features include:

• Quad Ported Fibre Channel Host Adapters

• Dual ported Fibre Channel drives for higher availability and reliability

• Quad Modular Fibre Channel RAID Controllers are designed to match the I/Odemands of 2+ GHz nodes.

• Fibre optic point-to-point connections between node and array provide for higherperformance and greater distances between node and array.


NCR 6841-2456 Fibre Channel Disk Array

What’s new?• Utilizes new drive tray that provides 2 Gbit access to disks.

I/O path is 2 Gbit from hosts to array controllers to diskdrives.

• Primary disk drive is 36 GB.

Cabinet characteristics:• Support for two arrays for a total of 112 drives.

Fibre Channel Disk Array characteristics:• Dual hot swappable redundant 4884 Fibre Channel RAID

controllers.• Utilizes 2 Gbit Fibre Channel 36 GB 15K rpm disk drives.• I/O maximum throughput (per array) is approximately 250

MB/sec.• Typical configuration usage within a clique:

– 4 nodes sharing 3 arrays; provides 750 MB/sec I/O bandwidth• Support for RAID 1 and RAID 5.

NCR

SMC - 2U

4884 Controllers

4884 Controllers

6841-2456



NCR 6841-6456 Fibre Channel Disk ArrayCabinet characteristics include:

• 2U Server Management Chassis

• Three 2U UPS with Dual AC Distribution Boxes

• Support for two arrays for a total of 112 drives and 8.2 TB raw data capacity using73 GB disk drives


• Dual hot swappable redundant 5884 Fibre Channel RAID controllers with 1 GB ofcache and an Intel Pentium III 850 MHz microprocessor.

• Each 6841-6456 disk array has support for up to 56 hot swappable 2 Gbit FibreChannel 73GB 15K rpm disk drives (Seagate Cheetah drives).

• Quad Fibre Channel Host Adapter (2Gb/sec) point-to-point connectivity betweenprocessing node and disk arrays.





• Quad Modular Fibre Channel RAID Controllers are designed to match the I/Odemands of 2+ GHz nodes.

• Fibre optic point-to-point connections between node and array provide for higherperformance and greater distances between node and array.


NCR 6841-6456 Fibre Channel Disk ArrayWhat’s new?

• Utilizes new and faster RAID controllers - 5884.• Utilizes new drive tray that provides 2 Gbit access to disks.

I/O path is 2 Gbit from hosts to array controllers to diskdrives.

• Primary disk drive is 73 GB.

Cabinet characteristics:• Support for two arrays for a total of 112 drives.

Fibre Channel Disk Array characteristics:• Dual hot swappable redundant 5884 Fibre Channel RAID

controllers.• Utilizes 2 Gbit Fibre Channel 73 GB 15K rpm disk drives.• I/O maximum throughput (per array) is approximately 300

MB/sec.• Typical configuration usage within a clique:

– 4 nodes sharing 2 arrays; provides 600 MB/sec I/O bandwidth• Support for RAID 1 and RAID 5.

NCR

SMC - 2U

5884 Controllers

5884 Controllers

6841-6456



5380 and 6841-2456 Disk ArraysThe facing page contains an example of a 4-node clique sharing three NCR 6841-2456 diskarrays.

Each SMP has 2 Quad Fibre Channel adapters (2 Gbit/sec.). Fibre Channel cables are point-to-point connections.


5380 and 6841-2456 Disk Arrays

Typical configuration with6841-2456 arrays is:• 4 node clique sharing 3

arrays - 160 drives.• Each node configured

with 10 AMPs.• Each AMP has a Vdisk

assigned with 4 drivesconfigured with RAID 1.

• Each AMP VdiskMaxPerm - 72 GB.

• 40 AMPs @ 72 GB -total MaxPerm in clique- 2.88 TB.

DAC-A DAC-B

6841-2456

DAC-A DAC-B

6841-2456

DAC-A DAC-B

6841-2456

0 4 36…….

SMP001-4 AMPs

1 5 37…….

SMP001-5 AMPs

2 6 38…….

SMP002-4 AMPs

3 7 39…….

SMP002-5 AMPs

RAID 1

Vdisk 0

Pdisk 0

Pdisk 1

MaxPerm 72 GB

AMP 0

36 GB

36 GB

36 GB

36 GB



5380 and 6841-6456 Disk ArraysThe facing page contains an example of a 4-node clique sharing two NCR 6841-6456 diskarrays.

Each SMP has 2 Quad Fibre Channel adapters (2 Gbit/sec.). Fibre Channel cables are point-to-point connections.


5380 and 6841-6456 Disk Arrays

0 4 24…….

SMP001-4 AMPs

1 5 25…….

SMP001-5 AMPs

2 6 26…….

SMP002-4 AMPs

3 7 27…….

SMP002-5 AMPs

DAC-A DAC-B

6841-6456

DAC-A DAC-B

6841-6456

Typical configuration with6841-6456 arrays is:• 4 node clique sharing 2

arrays - 112 drives.• Each node configured

with 7 AMPs.• Each AMP has a Vdisk

assigned with 4 drivesconfigured with RAID 1.

• Each AMP VdiskMaxPerm - 146 GB.

• 28 AMPs @ 146 GB -total MaxPerm in clique -4.08 TB.

RAID 1

Vdisk 0

Pdisk 0

Pdisk 1

73 GB

73 GB

73 GB

73 GBMaxPerm146 GB

AMP 0



Performance versus CapacityThe facing page compares the two array configurations.

For typical Teradata warehouse configurations, the 6841-2456 configuration provides theperformance needed and is the recommended configuration.


Performance versus Capacity

Assume two typical clique configurations:• Four 5380 nodes - three 6841-2456 arrays (performance configuration)• Four 5380 nodes - two 6841-6456 arrays (capacity configuration)

6841-2456 - Clique with four 5380 nodes sharing three 6841-2456 arrays• Effectively, 720 GB data per node (40 drives @ 36 GB x RAID 1)• Performance benefits

– Each AMP is controlling and accessing less data - 72 GB per AMP– I/O benefits include 6 controllers providing access to data and I/O bandwidth

across clique - 750 MB• Provides superior performance with typical Teradata workload and active

data warehouse environments.

6841-6456 - Clique with four 5380 nodes sharing two 6841-6456 arrays• Effectively, 1022 GB data per node (28 drives @ 73 GB x RAID 1)• Capacity benefit

– Each AMP is controlling and accessing more data - 146 GB per AMP• Most useful when creating data warehouses that require less expensive

storage and where performance is not as important.



SummaryThe facing page summarizes some important concepts regarding this module.


Summary

• The NCR 4950/5350 and 4980/5380 systems leverage a common hardwarebuilding block - the SMP node.

– The SMP chassis is 5U in height - allowing up to 4 SMPs in a standardrack.

– SMPs utilize 2.8 GHz CPUs.

– Expandable In-cabinet upgrade

• Add nodes to partially populated rack

• The NCR Enterprise Storage disk arrays are used to house rack mountablestorage components such as modular disk arrays.

– NCR 6841-2456 primarily uses 36 GB disk drives

– NCR 6841-6456 primarily uses 73 GB disk drives

• AWS provides a central point of operational control by utilizing the ServerManagement components in the system


Review Questions


1. List 3 improvements of the 5380 SMP as compared to the 5350 SMP.________________________________________________________________________________________________________________________________________________________________________________________________

2. List an improvement of the 6841-2456 as compared to previous 6840-1456 disk array.________________________________________________________________

3. List an improvement of the 6841-6456 as compared to 6841-2456 disk array.________________________________________________________________

Play the numbers games - match the number to a definition.

___ 1. 16 A. Typical # of AMPs in clique with 6841-2456 disk array___ 2. 28 B. Typical # of AMPs in clique with 6841-6456 disk array___ 3. 36 C. Max number of drives in one NCR enterprise storage disk array___ 4. 40 D. Max number of SMPs without using BYNET racks___ 5. 56 E. Typical disk drive size (GB) for 6841-6456 disk array___ 6. 73 F. Typical disk drive size (GB) for 6841-2456 disk array



Notes

How Teradata uses NCR Systems Page 13-1

Module 13


How Teradata uses NCR Systems


• Describe the fundamental relationship between UNIX, logicalunits, and disk array controllers.

• Describe the fundamental relationship between Vdisks,Pdisks, slices, and LUNs.

• Specify the maximum disk space that an AMP can manage.

• Given a new disk array and a set of configurationparameters, determine the number of LUNs and the size ofeach LUN as configured by PUT and pdeconfig.


Page 13-2 How Teradata uses NCR Systems

Notes


Table of Contents

TERADATA AND THE PROCESSING NODE................................................................................................. 13-4FSG CACHE......................................................................................................................................................... 13-4

SMP MEMORY AND TERADATA .................................................................................................................... 13-6SMP SHARED MEMORY ................................................................................................................................... 13-8MEMORY MANAGED BY THE OPERATING SYSTEM ............................................................................ 13-10FSG CACHE ........................................................................................................................................................ 13-12BUDDY NODE BACKUP (UNIX MP-RAS ONLY) ........................................................................................ 13-14VPROC NUMBER ASSIGNMENT................................................................................................................... 13-16VPROC NUMBER ASSIGNMENT EXAMPLE.............................................................................................. 13-18DISK ARRAYS FROM A O.S. PERSPECTIVE.............................................................................................. 13-20LOGICAL UNITS AND SLICES....................................................................................................................... 13-22LUN AND SLICING GUIDELINES.................................................................................................................. 13-24TERADATA, THE O.S., AND DISK ARRAYS ............................................................................................... 13-26TERADATA AND A 1456 DISK ARRAY (LOGICAL VIEW) ..................................................................... 13-28TERADATA AND A 1456 DISK ARRAY (DETAIL VIEW).......................................................................... 13-30LOGICAL UNITS AND VDISKS EXAMPLE ................................................................................................. 13-32

EXAMPLE OF 72 GB VDISK................................................................................................................................ 13-34TERADATA FILE SYSTEM CONCEPTS....................................................................................................... 13-36FULL CYLINDER READ .................................................................................................................................. 13-38PUT, PDECONFIG, AND DISK ARRAYS....................................................................................................... 13-40SUMMARY .......................................................................................................................................................... 13-42REVIEW QUESTIONS....................................................................................................................................... 13-44



Teradata and the Processing NodeEach AMP and PE vproc will be allocated 40 MB of memory. This example assumes 10AMPs and 2 PEs for a total of 12 vprocs in the SMP. Also note that there is one additionalinternal vproc that is allocated 40 MB of memory. This makes for a total of 13 vprocs formemory allocation.

The operating system will typically require 60 MB (MP-RAS) to 100 MB (Windows 2000)of memory for itself.

Practical experience (for most environments) indicates that the operating system (e.g.,UNIX) needs more than the 60 MB allocated during startup. Therefore, it is recommendedthat at least 100 MB to 300 MB of additional free memory be made available to theoperating system. This is accomplished by not giving 100% of the remaining memory toFSG. It is recommended that the FSG Cache Percent be set to a value less than 100%.

FSG Cache Percent = (FSG Cache – 200 MB) / FSG Cache

The default of 80% for FSG Cache Percent works well for most configurations.

FSG CacheFSG Cache is primarily used by the AMPs to access memory resident database segments.

When Teradata needs to read a database block, it checks FSG Cache first.

FSG is also used to backup AMP updates on other SMPs. This is referred to as “buddynode” backup.

Uses of FSG Cache

• Permanent data blocks• Cylinder Indexes• Spool data blocks• Transient Journals• Permanent Journals• Synchronized scan (sync scan) data blocks• Buddy Backup data blocks


Teradata and the Processing Node

UNIX SVR4 MP-RAS or Windows 2000

Teradata PDE (BaseVproc)Teradata TPA S/W

Process Control Memory Mgmt. I/O Mgmt. (Device Drivers)

MemoryCPUs

Pentium IV3.06 GHz

BIC4M QFC Eth.

BYNET DiskArrays

LANs

Pentium IV3.06 GHz

PDE - Parallel Data ExtensionsQFC - Quad Fibre Channel

PE vproc AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

AMPvproc

PE vproc

FSG (File Segment Cache) - managed by PDE



SMP Memory and TeradataThe example on the facing page illustrates an SMP with 2 GB of memory executing theTeradata.

Each AMP and PE vproc will be allocated 40 MB of memory. This example assumes 10AMPs and 2 PEs for a total of 12 vprocs in the SMP. Also note that there is one additionalinternal vproc that is allocated 40 MB of memory. This makes for a total of 13 vprocs formemory allocation.

FSG – File Segment Cache


SMP Memory and Teradata

20% of remaining space – 290 MB available to O.S.

FSG (File Segment Cache)

(Examples of use – Data Blocks & Cylinder Indexes)

Managed by PDE Software

80% of remaining space – 1158 MB available for FSG

PDE (BaseVproc) – 40 MB Hash MapsConfiguration Maps

40 MB

PEVproc

RTSD/D

Cache

40 MB

AMPVproc

MasterIndex

40 MB

PEVproc

RTSD/D

Cache

40 MB

AMPVproc

MasterIndex

40 MB

AMPVproc

MasterIndex

40 MB

AMPVproc

MasterIndex

40 MB

AMPVproc

MasterIndex

…..

Operating System – 60 to 80 MB

Ex. 2048 MB – 2 GB Memory

Examples of objects that arememory resident:

Hash MapsConfiguration MapsMaster IndexesRTS - Request-to-Steps CacheD/D - Data Dictionary Cache

Free MemoryO.S. 80 MB13 Vprocs @ 40 MB each

= 520 MB20% Remaining Space

= 290 MB2048 MB

– 80 MB – 520 MB – 290 MB

FSG Cache 1158 MB



SMP Shared MemoryThe example on the facing page illustrates an SMP with 2 GB of memory. This pagesummarizes memory use from the previous page. Each AMP and PE vproc will be allocated 40 MB of memory. This example assumes 10AMPs and 2 PEs for a total of 12 vprocs in the SMP. Also note that there is one additionalinternal vproc that is allocated 40 MB of memory. This makes for a total of 13 vprocs formemory allocation.


SMP Shared Memory

Xctl Parameter - FSG Cache Percent - usually set between 80 and 90%

Managed byPDE FSGsoftware.

Memorymanaged by

O.S.

When the operatingsystem boots, itallocates approximately60 to 80 MB for itself and40 MB for each vproc.

FSG - pool of memorymanaged by PDE andeach AMP uses what itneeds.

Ex. 2048 MB - 2 GBMemory

Operating System 80 MB

13 Vprocs @ 40 MB each 520 MB

FSG Cache

100% - 1448 MB

90% - 1303 MB

80% - 1158 MB



Memory Managed by the Operating SystemThe facing page lists examples of how free memory is used.

Practical experience (for most environments) indicates that UNIX needs more than the 60MB allocated during startup. Therefore, it is recommended that at least 100 MB to 300 MBof additional free memory be made available to UNIX. This is accomplished by not giving100% of the remaining memory to FSG. It is recommended that the FSG Cache Percent beset to a value less than 100%.

FSG Cache Percent = (FSG Cache – 200 MB) / FSG Cache

The default of 80% for FSG Cache Percent works well for most configurations.


Memory Managed by the Operating System

Memory managed and used by the operating system and the vprocs issometimes called “free memory”. Examples of how this memory is used:

• Vprocs for non-Teradata file system activity– AMP worker tasks– Parser and dispatcher tasks– Messages - communication between vprocs– Dictionary cache– Request to Steps cache– Temporary work space for vprocs

• Administrative programs such as:– program text and data– message buffers (ex., TCP/IP)– kernel resources– other applications (ex., FastLoad)



FSG CacheThe facing page lists examples of how FSG Cache is used.


FSG Cache

• FSG Cache is primarily used by the AMPs to access memory residentdatabase segments. When Teradata needs to read a database block, itchecks FSG Cache first.

• FSG is also used to backup AMP updates on other SMPs. This is referredto as “buddy node” backup.

• Uses of FSG Cache– Permanent data blocks– Cylinder Indexes– Spool data blocks– Transient Journals– Permanent Journals– Synchronized scan (sync scan) data blocks– Buddy Backup data blocks



Buddy Node Backup (UNIX MP-RAS only)The facing page discusses the concept of a Buddy Node.


Buddy Node Backup (UNIX MP-RAS only)

• A “buddy node” is another SMP in the same clique which is used tobackup AMP updates.

• Updates to data blocks and/or cylinder indexes are written immediatelyto disk based on following parameters:– Data Blocks only - DBSControl “WriteDBsToDisk” set to TRUE– Data Blocks & Cylinder Indexes - Xctl “Write All to Disk” set to ON

AMP

Data BlockFSG

Data BlockFSG

BYNET

Buddy Node

AMPSQL UPDATE



Vproc Number AssignmentThe facing page describes how Vprocs are assigned numbers.


Vproc Number Assignment

• Each PDE, AMP, and PE is assigned a unique Vproc # in the system.

• PDE Vproc #s (start at 16384 and increment by 1)– PDE on first SMP - 16384– PDE on second SMP - 16385– PDE on third SMP - 16386

• AMP Vproc #s (start at 0 and increment by 1)– First AMP - 0– Second AMP - 1– Third AMP - 2

• PE Vproc #s (start at 16383 and decrement by 1)– First PE - 16383– Second PE - 16382– Third PE - 16381

Appear in utilities suchas xgtwglobal.

Appear in utilities suchas vprocmanager,qryconfig, etc.

Appear in utilities suchas vprocmanager,qryconfig, etc.



Vproc Number Assignment ExampleThe facing page has an example of a 4-node system and vproc number assignments.


Vproc Number Assignment Example

SMP001-4PDE 16384

PE16383

AMP0

AMP4

AMP8

AMP12

AMP16

AMP20

AMP24

PE16382

SMP001-5PDE 16385

PE16381

AMP1

AMP5

AMP9

AMP13

AMP17

AMP21

AMP25

PE16380

SMP001-6PDE 16386

PE16379

AMP2

AMP6

AMP10

AMP14

AMP18

AMP22

AMP26

PE16378

SMP001-7PDE 16387

PE16377

AMP3

AMP7

AMP11

AMP15

AMP19

AMP23

AMP27

PE16376



Disk Arrays from a O.S. PerspectiveAn Operating System (e.g., UNIX MP_RAS) is used to reading and writing data to/from anindividual disk. Disk arrays trick the operating system into thinking it is writing to a singledisk. A disk array LUN looks to the operating system like a single disk. When theoperating system gets ready to do a read or a write, the disk array controller steps in andsays, “I’ll handle that for you”. The operating system says, “I am writing to a single diskand its address is c10t0d0s1”.

The operating system does not directly read or write to a disk in a disk array environment.The operating system communicates with the disk array controller. The operating systemactually reads or writes the data from a logical unit (often referred to as a LUN or aVolume). A logical unit (LUN) or Volume is a logical disk and not a physical disk.

The operating system does not know (or care) if a LUN or Volume is RAID 0, RAID 1, orRAID 5. The operating system does not know if the drive group is one disk, two disk, orfour disks. The operating system does not know if the data is spread across one disk or fourdisks. The operating system simply sees the logical unit as a single disk.

The standard operating system utilities that are used to manage, configure, and utilize aphysical disk are also used to manage, configure, and utilize a logical disk or LUN. WithTeradata, the PUT and pdeconfig utilities are used to configure the disk array.

The array controller performs the actual input/output operations to its disks. The arraycontroller is responsible for handling the different RAID technologies.


Disk Arrays from an O.S. Perspective

• A logical unit (LUN) or Volume is a single disk to UNIX.

• For UNIX systems, standard UNIX utilities that are used to slice physicaldisks are also used to slice LUNs.

• The operating system does not know or care about the specific RAIDtechnology being used for a LUN or Volume.

Operating System

LUN or

Volume

LUN or

Volume

LUN or

Volume

LUN or

Volume

The operating system (e.g., UNIX MP-RAS or Windows 2000) thinks it isreading and writing to 4 individual disks.



Logical Units and SlicesA logical unit (just like a physical disk) can be divided into multiple slices. Aslice is a portion of a logical unit. A slice is typically used in one of two ways.

• Used to hold a UNIX file system (e.g., ufs file system). • Provides a raw data storage area (raw disk slice) that can be used by

databases.

Through sysadm, osa, or command-line utilities, an administrator specifies thestarting physical block (sector #) for a slice and the length of slice (number ofphysical blocks).

A particular slice is accessed by the s_ portion of the device name/dev/rdsk/c_t_d_s_.

Slice 0 represents the entire logical unit or disk.

Slice 7 is the Boot slice and is 34 sectors or physical blocks. Slice 7 contains theVolume Table of Contents (VTOC) for the entire logical unit.

Acronym: FS – File System


Logical Units and Slices

With UNIX MP-RAS, a logical unit (LUN) will have a single UNIX partition whichis typically divided into multiple slices.

How are slices typically used?• Hold a UNIX file system (e.g., vxfs).• Provides raw data storage area (raw disk slice) for databases (e.g., Teradata).• Slice 7 (Boot Slice) contains VTOC which has starting location (physical block #)

and size of each slice on the disk.

0

7 - Boot

8 - Raw sliceassignedto a Pdisk

TeradataPdisk space

7 - Boot

1 - Root FS

f

2 - Swap 0UNIX MP-RASSystem Disk



LUN and Slicing GuidelinesThe facing page lists various LUN and slicing guidelines.

For Teradata (UNIX) V2, the maximum amount of space that one AMP can access is basedon the following calculation:

61000 logical cylinders x 1488 sectors/cylinder x 512 bytes/sector

which equals 46,473,216,000 bytes or approximately 43 GB where a GB is 10243

For Teradata (UNIX) V2R3, V2R4, and Teradata for NT, the maximum amount of spacethat one AMP can access is based on the following calculation:

61000 logical cylinders x 3872 sectors/cylinder x 512 bytes/sector

which equals 120,930,304,000 bytes or approximately 112 GB where a GB is 10243.

For Teradata V2R4.1 and V2R5.0, the maximum amount of space that one AMP can accessis based on the following calculation:

700,000 logical cylinders x 3872 sectors/cylinder x 512 bytes/sector

which equals 1,387,724,800,000 bytes or approximately 1.26 TB where a TB is 10244.

This increase to 700,000 cylinders will allow for the use of newer, higher capacity diskdrives with Teradata.


LUN and Slicing Guidelines

• Each LSI array has a maximum of 32 LUNs or Volumes.

• UNIX 3.01 and above (currently on UNIX 3.02)

– Maximum LUN size = 1 TB– Maximum slice size = 1 TB

• Teradata sizing issues (maximum space that one AMP can address)

– Teradata V2 Releases 1 and 2 – 43 GB

(61000 cylinders x 1488 sectors/cylinder x 512 bytes/sector)

– Teradata V2R3, V2R4, and Windows NT – 112 GB

(61000 cylinders x 3872 sectors/cylinder x 512 bytes/sector)

– Teradata V2R4.1, V2R5.0 – 1.26 TB

(700,000 cylinders x 3872 sectors/cylinder x 512 bytes/sector)



Teradata, the O.S., and Disk ArraysThe Teradata Relational Database Management System (RDBMS) has long been recognizedas one of the most effective database platforms for the storage and management of verylarge relational databases.

NCR’s Teradata database implementation executes as an application under UNIX on one ormore processing nodes. The two key pieces of software that make up the Teradata RDBMSare the PE software and the AMP software. These will be briefly described on the next fewpages.

Users access the Teradata RDBMS by issuing SQL commands - usually from channel-attached hosts or LAN attached workstations. The user request is handled by ChannelDriver or Gateway software and is passed to a Parsing Engine (PE) which processes theSQL request. PE software manages the user session, interprets (parses) the SQL request,creates an execution plan, and dispatches the steps of that plan to the AMP(s).

AMPs provide access to user data stored within tables that are physically stored on diskarrays.

Each AMP is associated with a Vdisk. Each AMP sees its Vdisk as a single disk. Teradata(AMP software) organizes its data on its disk space (Vdisk) using a Teradata “File System”structure. A “master index” is used to locate “cylinder indexes” which are used to locatedata blocks that contain data rows.

A Vdisk is actually composed of multiple slices (also called Pdisks - Physical disk) that arepart of a LUN (Logical Unit) in a disk array. UNIX and the array controllers work at theLUN level.

A logical unit (just like a physical disk) can be divided into multiple slices. A slice is aportion of a logical unit.

An AMP is assigned to a Vdisk. A Vdisk is composed of one or more Pdisks. A Pdisk is aUNIX slice on a LUN. Typically, slices numbered “8, 9, a, b” are assigned to Pdisks.

The PUT and pdeconfig utilities are used to define a Teradata configuration.


Teradata, the O.S., and Disk Arrays


Single Disk

PE

FileSystem

PDE

O.S. Logical Disks

User

Teradata Pdisk = UNIX Slice= Windows 2000 Partition

AMP

Pdisk 0

LUN 0

Pdisk 1

LUN 1

Pdisk 0

Pdisk 1

Vdisk



Teradata and a 1456 Disk Array (Logical View)The facing page illustrates the relationship between Teradata and a 6840-1456disk array subsystem.

Each virtual AMP is assigned to a virtual disk (Vdisk). AMP 0 is assigned toVdisk 0, AMP 1 is assigned to Vdisk 1, etc.

A Vdisk is actually composed of multiple slices (also called Pdisks - Physicaldisk) that are part of a LUN (Logical Unit) in a disk array.

A Vdisk effectively represents a set of disks in a disk array. In this example, aVdisk represents a rank of 4 disks in a disk array that is configured to use RAID 1technology.

If the disk array has 36 GB disks and RAID 1 protection is used, then one rank ofdisks (4 disks) has 72 GB of available user space.

4 disks x 36 GB x .50 (parity is 50%) = 72 GB

If the disk array has 36 GB disks and RAID 5 protection is used, then one rank ofdisks (4 disks) has 108 GB of available user space.

4 disks x 36 GB x .75 (parity is 25%) = 108 GB

Therefore, if the Vdisk is configured (assigned) to one rank of 36 GB disks(RAID 1), then the associated AMP has 72 GB of perm disk space available to it.If the disk array has 36 GB disks (RAID 5), then each AMP has 108 GB of permdisk space.

The configuration of LUNs/slices and the assignment Pdisks/Vdisks to AMPs isdone through the PUT and pdeconfig utilities.

Acronyms:DAC - Disk Array ControllerQFC - PCI Quad Fibre Channel adapter


Teradata and a 1456 Disk Array (Logical View)

6840-1456 Disk Array with 56 Disks - Logical View

AMPvproc

0

}

AMPvproc

2

}AMPvproc

4}

AMPvproc

6

}AMPvproc

8

}

AMPvproc

10

}

AMPvproc

12

}

SMP

AMPvproc

1

}

AMPvproc

3

}

AMPvproc

5

}

AMPvproc

7

}

AMPvproc

9

}

AMPvproc

11

}

AMPvproc

13

}

Vdisk0

Vdisk2

Vdisk4

Vdisk6

Vdisk8

Vdisk10

Vdisk12

Vdisk1

Vdisk3

Vdisk5

Vdisk7

Vdisk9

Vdisk11

Vdisk13

SMP



Teradata and a 1456 Disk Array (Detail View)The facing page continues a detailed view of the previous example.


Teradata and a 1456 Disk Array (Detail View)

DAC

DAC

SMP

QFC

QFC

Vdisk0

Vdisk2

Vdisk4

Vdisk6

Vdisk8

Vdisk10

Vdisk12

AMPvproc

0

AMPvproc

2

AMPvproc

4

AMPvproc

6

AMPvproc

8

AMPvproc

10

AMPvproc

12

QFC

QFC

Vdisk1

Vdisk3

Vdisk5

Vdisk7

Vdisk9

Vdisk11

Vdisk13

AMPvproc

1

AMPvproc

3

AMPvproc

5

AMPvproc

7

AMPvproc

9

AMPvproc

11

AMPvproc

13

6840-1456 Disk Array

SMP

QFC - Quad FibreChannel adapter



Logical Units and Vdisks ExampleThe facing page completes this example by showing the Logical Units and the associatedVdisks.


Logical Units and Vdisks Example

36 GB

36 GB

36 GB

36 GB

LUN 0

Pdisk 0

LUN 1

Pdisk 1

Vdisk 0

Pdisk 0

Pdisk 1

AMPvproc

0

72 GB ofMax

PERMspace

Assumes 36 GB Disks & RAID 1

36 GB

36 GB

36 GB

36 GB

LUN 2

Pdisk 2

LUN 3

Pdisk 3

Vdisk 1

Pdisk 2

Pdisk 3

AMPvproc

1

72 GB ofMax

PERMspace



Example of 72 GB VdiskThe paging page contains a typical example of a 72 GB Vdisk. It would contain36,653 cylinders; each cylinder is 3872 sectors in size. Therefore, a cylinder isapproximately 2 MB in size (3872 x 512 bytes).


Example of 72 GB Vdisk

Assumes RAID 1 Mirroring

Vdisk

Cylinder 0

1

2

3

36653

72 GB

Pdisk 1

Pdisk 0

36 GB

Cylinder 0

18327

36 GB

18328

36653

36 GB

Disk ID1,0

36 GB

Disk ID2,0

36 GB

Disk ID3,0

36 GB

Disk ID4,0

AMP

Physical DisksTeradata’s File Systemsoftware divides theVdisk into logicalcylinders; eachcylinder is 3872sectors in size.



Teradata File System ConceptsEach AMP has its own disk space managed by Teradata file system software. The filesystem software groups physical blocks into logical cylinders.

With Teradata V2R4.1, one AMP can manage up to 700,000 cylinders. Each cylinder has acylinder index (CI). Although an AMP can address this large number of cylinders, typicallyan AMP will only be responsible for a much smaller number of cylinders. For example, anAMP that manages 36 GB of disk space will have approximately 18,000 cylinders.

When an AMP is initialized (booted), it reads the Cylinder Indexes and creates an in-memory Master Index to the Cylinder Indexes.

Notes:

Teradata V2R2 – each Cylinder Index is 4 KB in size. The 4 KB Cylinder Index is usedwhen the cylinder contains 1488 sectors.

Teradata V2R3, V2R4, V2R4.1 – each Cylinder Index is 8 KB in size. The 8 KBCylinder Index is used when the cylinder contains 3872 sectors.

Teradata V2R5 – each Cylinder Index is 12 KB in size. The 12 KB Cylinder Index isneeded for the new file system structures with V2R5. The cylinder size is still3872 sectors.


Teradata File System Concepts

With V2R4 andV2R5, the cylindersize is 3872sectors.

For a 72 GB Vdisk,there will beapproximately36,650 cylinders.

Cylinder Index

Cylinder Index

Max # ofCylindersis approx.700,000

Data Block with rows




Cylinder1

Cylinder2

SMP MemoryMaster Index

Entry for CI #1Entry for CI #2

Entry for CI #700,000

Vdisk

Cylinder Index

Size of each Cylinder Index:V2R4.1 – 8 KBV2R5.0 – 12 KB



Full Cylinder ReadFull Cylinder Read allows retrieval operations to run more efficiently by reading a list ofcylinder-resident data blocks with a single I/O operation. This reduces I/O overhead fromonce per data block to once per cylinder.

A data block is a disk-resident structure that contains one or more rows from the same tableand is the smallest I/O unit for the Teradata file system. Data blocks are stored in physicaldisk sectors or segments, which are grouped in cylinders.

Full Cylinder Read improves the performance of systems with both fine-grained operationsand decision support workload. It eliminates the tradeoffs for short queries and concurrentupdates versus strategic queries.

Performance may benefit from Full Cylinder Read during operations such as:

• Full-table scan operations under conditions such as:• Large select• Merge insert/select• Merge delete• Aggregation:• Sum• Average• Minimum/Maximum• Count • Join operations that involve many data blocks, such as:• Merge joins• Product joins• Inner/outer joins

During installation, Full Cylinder Read is enabled by default with 4 slots per AMP. It isdisabled if FSG memory per AMP is below 36MB. The default number of cylslots is 4 perAMP.

• User can select between 2 and 40 cylslots per AMP.• Cylinder Read can be turned On and Off.• CR is automatically turned off when the memory is below 32MB per AMP.

Teradata Customer Support sets the CR flag to ON and uses MultiTool with Xctl/ctl tomodify the number of cylslots.

The typical node with 7 to 10 AMPS will continue to be configured with 2 GB of memorywith 5% of that memory set aside for cylinder slots. Memory allocated to cylinder slots canonly be used for cylinder reads. The benefit of cylinder reads is likely to outweigh thereduction in generic FSG cache.


Full Cylinder Read

The Full Cylinder Read feature is available with V2R5. Multiple pre-loadreads to a cylinder may be retrieved with a single large read, rather thanindividual reads.

~2 MBDataBlock

DB DB DB DB DB

CYLINDER

Enables efficient use of disk & CPU performance resources for the followingtable scan operations under specific conditions:

– aggregates: sum, avg, min, max, count– joins: merge joins, product joins, inner/outer joins:– large selects– merge delete– merge insert/select: empty or populated tables– full file update/deletes

Additional operations will use cylinder read in future releases.



PUT, Pdeconfig, and Disk ArraysThe PUT and pdeconfig utilities are used to configure a Teradata databaseenvironment. These utilities perform many functions in the configuration of aTeradata system. One of the key functions is to scan a system for disk arrays andconfigure those disk arrays for use with Teradata.

The chart on the facing page shows the PUT and pdeconfig defaultconfigurations for different disk arrays.

If the default configuration is not desired, then ACE can be used to manuallyconfigure an array.

Notes:

RAID 1 – Classic mirroring (no striping); commonly used with TeradataRAID 1 + 0 - Striped MirroringRAID 5 - commonly used with Teradata with 4 GB and 9 GB disk drives


PUT, pdeconfig, and Disk Arrays

• The PUT and pdeconfig utilities are used to configure a Teradata databaseenvironment.

• One of its key functions is to scan a system for disk arrays and configurethose disk arrays for use with Teradata.

• Depending on the type of array, RAID Level, size and number of disks, theseutilities will configure arrays differently.

• This chart lists configuration defaults.

ModelNo.

RAIDLevel

Max # ofDisks

Disks/Group

# ofGroups

LUNs/Group

TotalLUNs

Disk Size (GB) 18 36 73LUN Sizes (GB)

1440 1 40 2 20 1 20 18 36 731440 1 + 0 40 4 10 1 10 36 72 1461440 5 40 4 10 1 10 54 108 219

1452 1 52 2 26 1 26 18 36 731452 1 + 0 52 4 13 1 13 36 72 1461452 5 52 4 13 1 13 54 108 219

1456 1 56 2 28 1 28 18 36 731456 1 + 0 56 4 14 1 14 36 72 1461456 5 56 4 14 1 14 54 108 219


Summary

• Memory managed and used by the operating system and the vprocs issometimes called “free memory”.

• PDE software manages FSG Cache.

– FSG Cache is primarily used by the AMPs to access memory residentdatabase segments.

• The operating system and Teradata does not know or care about the RAIDtechnology being used.

• A LUN or Volume looks like a single disk to the operating system.

• With UNIX MP-RAS, a LUN or Volume is divided into slices and raw diskslices are assigned to Teradata Pdisks.

• With Windows 2000, a LUN or Volume is considered a partition and the rawpartition is assigned to a Teradata Pdisk.

• PUT and pdeconfig are utilities that are used to configure a Teradatadatabase environment.


Review Questions

1. Describe the fundamental relationship between UNIX, logical units, and disk array controllers.________________________________________________________________________________________________________________________________________________________________

2. Describe the fundamental relationship between Vdisks, Pdisks, slices, and LUNs.________________________________________________________________________________________________________________________________________________________________

3. Specify the maximum disk space that an AMP can manage.

Teradata V2R3 - ________________Teradata V2R5.0 - ________________

4. Given a new disk array and the following parameters, complete the following chart as configured bypdeconfig.

RAID # Disk Total # Size ofLevel of Disks Size of LUNs each LUN

RAID 1 40 18 GBRAID 1 56 36 GBRAID 5 40 18 GBRAID 5 56 36 GB



Notes

Acronyms Page A-1

Appendix A


Appendix A: Acronyms

This Appendix contains a listing of variousNCR and Teradata acronyms.

Page A-2 Acronyms

Notes

Acronyms Page A-3

-A-ABF Adaptive Bandwidth Feature ABI Application Binary Interface ABIOS Advanced BIOS ABM Asynchronous Balance Mode ABRD Automatic Baud Rate Detection AC Application Component, Alternating CurrentACE Array Configuration Editor ACF Advanced Communications Function ACF/NCP Advanced Communication Function for the Network Control Program ACF/SSP ACF/Software Support Program ACF/VTAM Advanced Communication Function for the Virtual Telecommunications Acces

Method ACID Atomicity, Consistency, Isolation, DurabilityADAM Application Data Access ManagerADCOM Advanced Distributed Communications System ADCS Advanced Data Communication System ASCII American Standard Code for Information InterchangeADE Application Development EnvironmentAFMS Advanced Function Management System AFMS/NDP Advanced Function Management System/Normalized Data Protocol AFMS/SEF Advanced Function Management System/System Exchange Format AI Artificial IntelligenceANSWER Alpha-Numeric Single Wire Electronic Recorder AOD Application Output Definition AOE Application Operating Environment AOS Automated Order System AP Application ProcessorAPI Application Programming InterfaceAMP Access Module ProcessorAPPC Advanced Program to Program CommunicationAS Administration StationASF Archive Storage FacilityASIC Application Specific Integrated CircuitATM Asynchronous Transfer Module (Networking)

Automated Teller Machine AUI Attachment Unit InterfaceAWS Administration WorkStation

Page A-4 Acronyms

-B-BAM BYNET Administrative MenusBASIC Beginner's All-purpose Symbolic Instruction Code BCD Binary-Coded Decimal BDL BYNET Data LinkBIC BYNET Interface Card (Adapter)BIOS Basic Input/Output System BIST Built-In Self TestBIU Battery Interface UnitBLIP BYNET Low Level Internet ProtocolBLLI BYNET Low Latency InterfaceBLM BYNET Link ManagerBMCA BYNET Micro Channel Architecture adapterBNC Bayonet-Neill-Concelman (Ethernet connector)BOOTP BOOTstrap Protocol BPCI BYNET Peripheral Component Interconnect bps bits per second BTEQ Batch/basic TEradata Query facilityBTAM Basic Telecommunications Access Method BYA4G BYNET A Switch V2 – 4 ports – Gigahertz speed (60 MB/sec.)BYA4M BYNET A Switch V2 – 4 ports – Gigahertz speed (60 MB/sec.)BYA4P BYNET A Switch V1 – 4 ports – 10 MB/sec.BYA16 BYNET A Switch V1 – 16 ports – 10 MB/sec.BYA16G BYNET A Switch V2 – 16 ports – Gigahertz speed (60 MB/sec.)BYA64GX BYNET A Switch V2 – 64 ports – Expandable - Gigahertz speedBYAS Component name for BYA16XS boardsBYB32 BYNET B Switch – 10 MB/sec.BYB64G BYNET B Switch V2 – 64 ports - Gigahertz speedBYNET Banyan Network - High speed interconnect

Acronyms Page A-5

-C-CA Cable AdapterCAD/CAM Computer-Aided Design/Computer-Aided Manufacturing CAE Computer Aided EngineeringCAI Century Analysis, Inc., Computer-Aided Instruction

Computer-Assisted Instruction CASE Computer Aided Software EngineeringCAT Configuration and TestCC Cluster Controller, Common Cable, Common Carrier CCB Change Control Board, Character Control Block, Command Control Block CCC Common Carrier Communication CCDLC Common Carrier Data Link Control CCITT Consultative Committee on International Telephone and TelegraphCD Compact DiskCD-ROM Compact Disk - Read Only MemoryCICS Customer Information Control SystemCFM Configuration File ManagementCLAN Cabinet Local Area NetworkCLC Cabinet Level Controller (5100)CLCX Cabinet Level Controller eXtended (5100)CLI Call Level InterfaceCM Communications ManagerCMB Chassis Management BoardCMIC Cabinet Management Interface Controller (for 5100)CMIC2 Chassis Management Interface Controller - 2 (for 4700/5150)CMOS Complimentary Metal Oxide Semiconductor COBOL Common Business-Oriented Language COP Communications Processors COS Corporate Office ServerCPU Central Processing Unit CR/LF Carriage Return/Line Feed CRC Cyclic Redundancy Check CRU Customer Replaceable UnitCSF Customer Support FacilityCSI Complex Service InterfaceCTG Channel TailgateCTS Clear to Send

Page A-6 Acronyms

-D-DA Disk ArrayDAC Disk Array ControllerDAMC Disk Array Module CabinetDARDAC Dual Active Redundant Disk Array ControllerDASD Direct-Access Storage DeviceDAT Digital Audio Tape, Dynamic Address TranslationDBA Database AdministratorDBC Database Computer (Teradata)DBLT Disk Bay Locator TableDBMS Database Management SystemDBS Database SubsystemDC Data Communications, Direct CurrentDCE Distributed Computing EnvironmentDCL Data Control LanguageDCRAM Disk Cache Random Access MemoryDD Derived DataDDE Direct Data ExchangeDDI Device Driver InterfaceDDL Data Definition Language DES Data Encryption StandardDIF Document Interchange Facility, Data Interchange FormatDIHA Diagnostic Interconnect Host AdapterDKI Driver Kernel InterfaceDIM Diagnostic Interconnect ModuleDIMM Dual In-line Memory ModuleDIP Database Initialization Procedure DISCO Disconnect memoryDLPI Data Link Provider InterfaceDMA Direct Memory Access DMBI Distributed Management Boards InterfaceDMCA Dual Micro Channel ArchitectureDML Data Manipulation Language DNS Domain Name Server DOD Department Of DefenseDOS Disk Operating System DP Diagnostic ProcessorDR Defect ReportDRAM Dynamic Random Access MemoryDS Desk SideDSC Data Stream CompatibleDSS Decision Support SystemDSU Disk Storage UnitDTP Distributed Transaction ProcessingDTR Data Terminal Ready

Acronyms Page A-7

-E-EA EaseAdvantageEAFW EaseAdvantage FrameworkEBCA EISA Bus Channel AdapterEBCDIC Extended Binary Coded Decimal Interchange CodeECC Error Check and CorrectionECL Emitter Coupler LogicEDAC Error Detection and CorrectionEDI Electronic Data InterchangeEEPROM Electronically Erasable Programmable Read Only MemoryEGA Enhanced Graphics Adapter (IBM)EISA Extended Industry Standard ArchitectureEOE Enhanced Operating EnvironmentEPROM Erasable Programmable Read-Only Memory ESM Environmental Services Monitor

-F-4GL Fourth Generation Language FAX FacsimileFC Fibre Channel FDDI Fiber Distributed Data InterfaceFE Field Engineer FEU Front End UnitFIFO First In, First Out FIPS Federal Information Processing Standards FK Foreign KeyFMLI Forms and Menu Language InterpreterFORTRAN FORmula TRANslationRU Field Replaceable UnitFTAM File Transfer and Access Method (OSI)FTP File Transfer ProtocolFUD Fear, Uncertainty, Doubt FW FirmwareFYI For Your Information

-G-GB Gigabyte (one billion bytes)GCA General Customer AvailabilityGOSIP Government Open Systems Interconnection ProfileGSC Global Support CenterGUI Graphical User Interface

Page A-8 Acronyms

-H-HASP Houston Automatic Spooling Program HATP High Availability Transaction ProcessingHCA Host Channel AdapterHDLC High-level Data Link ControlHW HardwareHz Hertz - cycles per second

-I-iABI Intel Application Binary InterfaceI2C or I2C Inter-Integrated CircuitIC Integrated Circuit ICI Initial Certified Installation ICMB Intelligent Chassis Management BusICSI Integrated Complex Service InterfaceIE Information EngineeringIEEE Institute of Electrical and Electronics EngineersI/L Memory Inter-Leave controllerIMS Information Management System, Inventory Management SystemIO Input/Output IP Internet Protocol IPMB Intelligent Platform Management BusIPS Integrated Peripheral SubsystemIRQ Interrupt Request ISA Industry-Standard Architecture ISAM Indexed Sequential Access MethodISD Interactive System DefinitionISDN Integrated Services Digital Network ISO International Standards OrganizationISP Internet Support Package ISPF Interactive System Productivity Facility ISV Independent Software VendorIT Information Technology

-J-JBOD Just a Bunch of DisksJCL Job Control Language JDO Joint Development Operation (NCR and Teradata)JES Job Entry Subsystem JFS Journaling File SystemJIT Just-In-Time

Acronyms Page A-9

-K-Kb KilobitKB Kilobyte (1,024 bytes)Kbaud Kilo BaudKbps Kilobits per secondKBPS Kilobytes per secondKHz KilohertzKSR Keyboard Send/Receive

-L-LAN Local Area NetworkLARC Limited Address Range CacheLCMP Loosely Coupled MultiprocessingLCD Liquid Crystal Display LDM Limited Distance Modem, Local Domain Manager LED Light Emitting DiodeLFM Log File ManagementLM Local Media ModuleLPB Local Peripheral BoardLPI Language Processors Inc.lpm lines per minute (or line-per-minute)LSU Logical Storage UnitLUN Logical Unit (disk array logical unit)

Page A-10 Acronyms

-M-MA Modular ArrayMAC Media Access Control, Medium Access Control, Message Authentication Code MAP Manufacturing Automation Protocol, Maintenance Analysis

Procedures, Master Application ProcessorMb Megabit MB Megabyte Mbps Megabits per second MLAN Management LANMCA Micro Channel ArchitectureMCCA Micro Channel to Channel AdapterMCIA Micro Channel Interface ArchitectureMHz MegaHertz - million cycles per secondMIPS Million Instructions Per SecondMIS Management Information System MO Method of Operation MPEG Moving Picture Experts Group MPP Massively Parallel ProcessingMSU Memory Storage UnitMTBDL Mean Time Between Data LossMTBF Mean Time Between FailuresMTBR Mean Time Between Repairs MTDP Micro Teradata Director ProgramMVS Multiple Virtual Storage (IBM mainframe OS)

-N-NA Network AgentNCMF NCR Communications Management Facility ND No DuplicatesNetBIOS Network Basic Input/Output SystemNFS Network File SystemNI Network InterfaceNIST National Institute of Standards and TechnologyNLQ Near Letter Quality NN No NullsNPPI Non-Partitioned Primary IndexNRZ Non-Return to Zero Ns Nanosecond NSC NCR Storage CabinetNTOS NCR Teradata Operating SystemNUPI Non-Unique Primary IndexNUSI Non-Unique Secondary IndexNVRAM Non-Volatile Random Access Memory (read and write memory)

Acronyms Page A-11

-O-OCC Open Cooperative ComputingOCCA Open Cooperative Computing ArchitectureOCR Optical Character Reader, Optical Character Recognition ODBC Open Database ConnectivityODS Operational Data StoreOE Operating EnvironmentOEM Original Equipment ManufacturerOLAP On-Line Analytical ProcessingOLCP On-Line Complex ProcessingOLE Object Linking and Embedding OLTP On-Line Transaction ProcessingOMC ORION Memory Controller chipOMC-DC ORION Memory Controller - DRAM ControllerOMC -DP ORION Memory Controller - Data Path ControllerONE Open Network EnvironmentONS Open Networking SystemOOP Object Oriented ProgrammingOPB ORION PCI Bridge chipOPS Oracle Parallel ServerOS Operating System OSA Open Systems ArchitectureOSF Open Software FoundationOSI Open Systems InterconnectOSI/DTP Open Systems Interconnect/Distributed Transaction ProcessingOSS Open System StandardsOTOS Open Teradata Operating System

Page A-12 Acronyms

-P-PBX Private Branch Exchange PC Personal Computer, Printed Circuit PCEB PCI EISA Bridge chipPCI Peripheral Component InterconnectPCMCIA Personal Computer Memory Card International AssociationPC/RA Probable Cause/Recommended ActionPCS Personal Computer SupportPDB Parallel Data BusPDE Parallel Database ExtensionsPDN Public or Private Data Network PE Parsing EnginePEP Parsing Engine ProcessorPK Primary KeyPL/1 Programming Language I PLAN Private LANPOPS Parallel Object Processing SystemPOS Point of Sale (noun), Point-of-Sale (terminal) (adj.)POSIX Portable Operating System Interface for Computing EnvironmentsPOST Power On System TestPPI Partitioned Primary IndexPQS PCI Quad SCSIPSC Power Shelf ControllerPSI Power Supply InterfacePTB Pass Through Board

-Q-QA Quality AssuranceQFC Quad Fibre ChannelQIC Quarter Inch Cartridge

Acronyms Page A-13

-R-RAID Redundant Array of Independent (formerly, Inexpensive) DisksRAM Random Access Memory RASUI Reliability, Availability, Serviceability, Usability, Installability RC Remote ClientRCMB Rack (or Remote) Chassis Management BoardRDAC Redundant Disk Array Controller (software)RDBMS Relational Database Management SystemRFC Request for Change RFP Request for Proposal RFS Remote File SystemRGB Red, Green, Blue RISC Reduced Instruction Set ComputingRJE Remote Job Entry RLM Revision Level ManagerRM Resource Manager (TOP END) or Rack MountROM Read Only MemoryROSE Remote Operation Service ElementRPC Remote Procedure CallRPG Report Program Generator rpm revolutions (or rotations) per minute RSC Remote System CallRTS Ready to Send, Request to SendRTS-CTS Request to Send - Clear to Send

Page A-14 Acronyms

-S-SAR System Activity ReportSCO Santa Cruz Operation UnixSCSI Standard (or Small) Computer System InterfaceSDL Screen Definition LanguageSDW Scalable Data WarehouseSE Systems Engineer SEEPROM Serial Electronically Erasable Programmable Read-Only MemorySI Self-Instruction SIMM Single In-line Memory ModuleSKU Stockkeeping Unit SLAN System Local Area NetworkSLC Second Level CacheSMB Server Management BoardSMC Server Management ChassisSMP Symmetrical Multi-ProcessingSNA Systems Network ArchitectureSNAG SNA GatewaySNO SuperNodeSNP Super Node ProcessorSOV Single Operational ViewSPI System Performance InvestigatorSQL Structured Query LanguageSQLCA Structured Query Language Communication AreaSQLDA Structured Query Language Data AreaSRAM Static Random Access MemorySS Support SentinelSSA Symbolic Service Address, System-to-System Adapter SSI Single System Image, Standard Serial Interface SSP Software Support Program, System Support Program (IBM)STMP Simple Mail Transfer Protocol SUS Startup SubsystemSVID System V Interface DefinitionSVR4 System V Release 4SW SoftwareSYSGEN System Generation

Acronyms Page A-15

-T-3GL Third Generation Language TAXI Transparent Asynchronous Xceiver InterfaceTBA To Be Announced TBD To Be Determined TCAM Telecommunications Access Method TCP/IP Transmission Control Protocol/Internet ProtocolTFTP Trivial File Transfer ProtocolTLI Transport Layer InterfaceTM Transaction ManagerTOD Time of Day (noun) or Time-of-Day (adj.)TOS Teradata Operating SystemTP Transaction ProcessingTPA Trusted Parallel Application or ArchitectureTPD Teradata Parallel Databasetps transactions per second TSO Time-Sharing Option TTL Transistor Transistor Logic TTU Teradata Tools and UtilitiesTTY Teletype (serial communication protocol)

-U-U Unit of Measurement - 1.75”UDP Unreliable Data ProtocolUFS UNIX File System UMB Universal Management BoardUPI Unique Primary IndexUPS Uninterruptable Power SourceUPS-IS Uninterruptable Power Source – Input SelectorUSI Unique Secondary Index

-V-VAR Value Added ResellerVDC Volts Direct CurrentVDT Video Display Terminal VGA Video Graphics Array VLF Very Low Frequency VLSI Very Large Scale Integration VM Virtual Machine VPIX "DOS under UNIX"VPROC Virtual ProcessorVSAM Virtual System Access MethodVTAM Virtual Telecommunications Access Method

Page A-16 Acronyms

-W-WAN Wide Area NetworkWinDDI Windows Data Dictionary Interface (application)WIN-TCP Wollongong Integrated Networking/Transmission Control Protocol

WORM Write Once - Read Many Optical DiskWTO Write to OperatorWYSIWYG What You See Is What You Get

-X-XA Extended ArchitectureXMIT Transmit

-Y-YMCA Ynet Micro Channel Architecture adapterYTD Year-to-Date

Book 1 - Answers to Review Questions Page B-1

Appendix B


Appendix B: Answers to Review Questions

This Appendix contains answers to thereview questions at the ends of the

modules forTeradata Factory Book #1

Page B-2 Book 1 - Answers to Review Questions

Notes


Module 1: Review Question Answers

1. Name the two primary operating systems that the Teradata RDBMS executes on.UNIX MP-RASWindows 2000

2. Which of the following represents a trillion bytes or a TB of data? ____

a. 106

b. 109

c. 1012

d. 1015

3. Which feature allows Teradata to process enormous volumes of data quickly? ____

a. High availability software and hardware componentsb. Parallelismc. Proven Scalabilityd. High performance servers from Intel

4. The Teradata RBDMS is primary a ____ .

a. Serverb. Client



Match each term with its definition below:

_f_ 1. Database

_e_ 2. Table

_b_ 3. Relational database

_a_ 4. Primary Key

_d_ 5. Null

_c_ 6. Foreign Key

a - A set of columns which uniquely identify a rowb - A set of logically related tablesc - One or more columns that are a PK somewhere in the databased - The absence of a valuee - A two-dimensional array of rows and columnsf - A collection of permanently stored data


M o d u le 3 : R ev iew Q u es tio n A n sw e rs

1 . W h at la n g u a g e is u se d to a c ce ss a T era d a ta ta b le ?

2 . W h at a re five T era d a ta d a ta b a se o b je c ts?

3 . W h at a re fo u r m a jo r c o m p o n e n ts o f th e T e rad ata a rc h ite c tu re ?

4 . W h at a re v ie w s?

5 . W h at a re m a cro s?

S Q L

T ab le s , v iew s , m a cro s , trig g e rs , a n d s to re d p ro c e d u re s

P E , A M P s, V d is ks , a n d M e ss ag e P a ss in g L ay er

- S u b s et o f ro w s an d co lu m n s o r o n e o r m o re ta b le s- V irtu a l ta b les- W in d o w in to o n e o r m o re tab le s

P red efin e d , s to re d s e t o f S Q L s ta te m en ts



1. What are the two software elements that accompany an application on all client side environments?

2. What is the purpose of the PE?

3. What is the purpose of the AMP?

4. How many sessions can a PE support?

Match Quiz __i_ 1. CLI __f_ 2. MTDP __e_ 3. MOSI __b_ 4. Parser __a_ 5. AMP __c_ 6. Message Passing Layer __d_ 7. TDP __g_ 8. Optimizer __h_ 9. Dispatcher __j__10. Parallelism

a. Does Aggregating and Lockingb. Validates SQL syntaxc. Connects AMPs and PEsd. Balances sessions across PEse. Provides Client side OS independencef. Library of Session Management Routinesg. PE S/W turns SQL into AMP stepsh. PE S/W sends plan steps to AMPi. Library of Teradata Service Routinesj. Foundation of Teradata architecture

CLI/ODBC and TDP/MTDP

Parse, optimize, and dispatch queries

Manage and retrieve data from disk storage

120



Complete the following.

1. Each AMP has its own memory and manages its own disk space and executes independently ofother AMPs. This is referred to as a shared nothing architecture.

2. The software component that allows Teradata to execute in different operating systemenvironments is the PDE.

3. A physical message passing interconnect is called the BYNET.

4. A clique provides protection from a node failure.

5. If a node fails, all vprocs will migrate to the remaining nodes in the clique. This feature is referredto as vproc migration.

6. The AWS provides a single point of operational control for NCR MPP systems.

7. List two non-NCR platforms that Teradata is supported on. Dell HP Compaq



True or FalseFalse 1. A database will always have tables.True 2. A user will always have a password.False 3. A user creating a subordinate user must give up some of his/her Perm Space.False 4. Creating tables requires the definition of at least 1 column and the user assignment of a Primary

Index.True 5. The sum of all user and database Perm Space will equal the total space on the system.False 6. The sum of all user and database Spool Space will equal the total space on the system.True 7. Before a user can read a table, a table SELECT privilege must exist in the DD/D for that user.False 8. Deleting a macro from a database reclaims Perm Space for the database.

9. Which of the following is FALSE about PERM space? ____

a. PERM space can be dynamically modified.b. The per/AMP limit of PERM space can not be exceeded.c. Maximum PERM space can be defined at the database or table level.d. Tables, index subtables, and stored procedures use PERM space.

10. Which of the following is FALSE about SPOOL space? ____

a. SPOOL space can be dynamically modified.b. The per/AMP limit of SPOOL space can not be exceeded.c. Maximum SPOOL space can be defined at the database or user level.d. Maximum SPOOL space can be defined at a value greater than the immediate parent's value.




For each statement, indicate whether it applies to:

UPI’s, NUPI’s, or Either

Either 1. Specified in CREATE TABLE statementUPI 2. Provides uniform distribution via the hashing algorithmEither 3. May be up to 64 columns in V2R5Either 4. Always a one-AMP operationUPI 5. Access will return (at most) a single rowEither 6. Used to assign a row to a specific AMPEither 7. Allows a null or nullsEither 8. Required on every tableNUPI 9. Permits duplicate rowsUPI 10. Used as a Primary Key implementation



Fill in the Blanks

1. The output of the hashing algorithm is called the Row Hash.

2. To determine the target AMP, the Message Passing Layer must lookup an entry in theHash Map based on the DSW or bucket number.

3. Two different PI values which hash to the same values are called Hash Synonyms .

4. A Row ID consists of a row hash plus a uniqueness value.

5. A uniqueness value is required to produce a unique Row ID because of hash synonymsand NUPI duplicates .

6. Once the target AMP has been determined for a PI search, the Master Index for that AMPmust be consulted.

7. The Cylinder Index points us to the address and length of the data block .



USIAccess FTS

# AMPs

# rows

Parallel Operation

Uses Hash Maps

Uses Separate Sub-table

Reads all data blocks of table

Fill each box with either Yes, No, or the appropriatenumber.

NUSIAccess

2 All All

0 - 1 0 - n 0 - all

N Y Y

Y N N

Y Y N

N N Y



Match the item to a lettered description.

a.) Provides for TXN rollback in case of failureb.) Open Teradata Backup applicationc.) Protects all rows of a tabled.) Logs changed rows for down AMPe.) Provides for recovery to a point in time f.) Applies to all tables and views withing.) Multi-platform archive utilityh.) Lowest level of protection granularity i.) Protects tables from AMP failure j.) Protects database from a physical drive failurek.) Group of AMPs used by Fallback

__f__ 1.) Database locks__c__ 2.) Table locks__h__ 3.) Row Hash locks__i__ 4.) FALLBACK__k__ 5.) Cluster__d__ 6.) Recovery journal__a__ 7.) Transient journal__g__ 8.) ARC__b__ 9.) NetBackup/NetVault__e__ 10.) Permanent journal__j__ 11.) Disk Array




1. Specify 2 improvements of the 53xx systems as compared to the 52xx systems.

Faster CPUs4 SMPs in a rack

2. What does the acronym represent and briefly define the purpose of the following subsystems?

SMP Symmetric Multi-Processor or Processing;provides general purpose processing capabilities

SMC Server Management Chassis; provides interface between AWS and rack-based cabinets

BYNET Banyan Network; high speed interconnect for data transfer/message passing between SMPs

AWS Administration Workstation; single point of operational control for multiple SMPs.

3. List the two LANs that connect the AWS to a 49xx/53xx system and briefly describe the purpose ofeach LAN.SLAN – (System LAN) connects AWS to CMICs for cabinet management, hardware faults, and consoleconnections.

PvtLAN – (Private LAN) connects AWS to SMPs for LAN connections, AWS status information, andsoftware faults.

4. Specify the names of the SMPs in 5380 SMP cabinet #3.SMP003-4 SMP003-5 SMP003-6 SMP003-7




1. List 3 improvements of the 5380 SMP as compared to the 5350 SMP.Faster CPU - 3.06 GHzFaster Front Side Bus - 533 MHzFaster PCI I/O Bus - 100 MHz

2. List an improvement of the 6841-2456 as compared to previous 6840-1456 disk array.Uses faster drive trays - 2 Gbit interface to disks.

3. List an improvement of the 6841-6456 as compared to 6841-2456 disk array. Uses faster disk array controllers - 300 MB of data I/O to SMPs.

Play the numbers games - match the number to a definition.

D 1. 16 A. Typical # of AMPs in clique with 6841-2456 disk array B 2. 28 B. Typical # of AMPs in clique with 6841-6456 disk array F 3. 36 C. Max number of drives in one NCR enterprise storage disk array A 4. 40 D. Max number of SMPs without using BYNET racks C 5. 56 E. Typical disk drive size (GB) for 6841-6456 disk array E 6. 73 F. Typical disk drive size (GB) for 6841-2456 disk array



1. Describe the fundamental relationship between the O.S., logical units, and disk array controllers.Disk array controllers manage a set of logical units (implemented across a set of disks) and makethese available to the operating system.

2. Describe the fundamental relationship between Vdisks, Pdisks, and LUNs or Volumes. Each AMP is assigned to a Vdisk which is made up of 1 or more Pdisks. A Pdisk is assigned to alogical unit or volume (in a disk array).

3. Specify the maximum disk space that an AMP can manage.

Teradata V2R3 - 112 GB Teradata V2R5.0 - 1.26 TB

4. Given a new disk array and the following parameters, complete the following chart as configured bypdeconfig.

RAID # Disk Total # Size ofLevel of Disks Size of LUNs each LUN

RAID 1 56 36 GB 28 36RAID 1 56 73 GB 28 73RAID 5 40 18 GB 10 54RAID 5 56 36 GB 14 108


Notes

Book 1 - Miscellaneous Platform Details Page C-1

Appendix C


Appendix C: Miscellaneous Platform Details

This Appendix contains details on ...

– NCR 48xx/52xx Systems– BYNET– AWS– Disk Arrays

Page C-2 Book 1 - Miscellaneous Platform Details

Table of Contents

COMPARING WORLDMARK SMP ARCHITECTURES...............................................................................C-44700 AND 5150 WORLDMARK SYSTEMS .......................................................................................................C-64800 AND 5200 WORLDMARK SYSTEMS .......................................................................................................C-8BYNET V2 SWITCH RACKS (CABINETS).....................................................................................................C-10

EXAMPLE 1: WORLDMARK 4855 – 4 NODES ......................................................................................................C-12EXAMPLE 2: WORLDMARK 4900 – 4 NODES ......................................................................................................C-14EXAMPLE 3: 5300 SYSTEM - 16 NODES ..............................................................................................................C-16EXAMPLE 4: 5300 SYSTEM - 32 NODES ..............................................................................................................C-18

RACK (CABINET) SUBSYSTEMS....................................................................................................................C-204800/4850/5200/5250 SMP ARCHITECTURE (LOGICAL VIEW)................................................................C-224851/4855/5251/5255 SMP ARCHITECTURE (LOGICAL VIEW)................................................................C-244900/5300 SMP ARCHITECTURE (LOGICAL VIEW) ..................................................................................C-26MEMORY..............................................................................................................................................................C-28

4800/4850/5200/5250 MEMORY.........................................................................................................................C-284851/4855/5251/5255 MEMORY.........................................................................................................................C-284900/5300 MEMORY...........................................................................................................................................C-28

WORLDMARK PCI I/O ADAPTERS................................................................................................................C-30SMP LOCAL MEDIA AND DISKS....................................................................................................................C-32

REMOVABLE MEDIA............................................................................................................................................C-32DISK DRIVES AND SCSI BACKPLANE..................................................................................................................C-32UNIX SCSI DEVICE NAMES...............................................................................................................................C-32

SUMMARY OF 4855/5255 SYSTEM IMPROVEMENTS ...............................................................................C-34SUMMARY OF 4900/5300 SYSTEM IMPROVEMENTS ...............................................................................C-36

SUMMARY OF 4900/5300 SMP IMPROVEMENTS .................................................................................................C-38WHAT IS BYNET VERSION 2.1? .....................................................................................................................C-40BYNET INTERFACE CARDS (BIC) OR ADAPTERS....................................................................................C-42

BIC2G BYNET ADAPTER..................................................................................................................................C-42BIC4G BYNET ADAPTER..................................................................................................................................C-42

BYNET SWITCHES.............................................................................................................................................C-44BYNET SWITCHES – BYA4M AND BYA16G ................................................................................................C-46

4700/5150 BYNET SWITCHES (NOT SHOWN) .....................................................................................................C-46BYNET SWITCHES (BYA64GX AND BYB64G).............................................................................................C-48

5300 BYNET V2 EXAMPLE – 128 NODES .........................................................................................................C-50BYNET V2 SWITCH CABINETS ..........................................................................................................................C-52BYNET V2 SWITCHES .......................................................................................................................................C-54EXAMPLE OF BYNET V2 SWITCHES - 128 NODES .............................................................................................C-56

BYNET™ LOGS AND UTILITIES....................................................................................................................C-58SMP DIAGNOSTIC UTILITIES...............................................................................................................................C-58


Table of Contents (cont.)

AWS CONNECTIVITY TO 48XX/52XX SYSTEM......................................................................................... C-60SYSTEM LAN (SLAN) ....................................................................................................................................... C-60PRIVATE LAN (PVTLAN) .................................................................................................................................. C-60AWS SYSTEM DISPLAY – LARGER CONFIGURATION.......................................................................................... C-62AWS MENU SELECTIONS ................................................................................................................................... C-64WINDOWS 2000 AWS MENU SELECTIONS ......................................................................................................... C-66

COMMAND MENU............................................................................................................................................. C-68CONNECT (TO PROCESSOR) ................................................................................................................................. C-68CONNECT TO AN SMP PROCESSOR ..................................................................................................................... C-70CONNECT TO DB WINDOW ................................................................................................................................. C-72EXIT - EXITING XCON ......................................................................................................................................... C-72

AWS EVENT LOG............................................................................................................................................... C-74CIRCULAR FILE ................................................................................................................................................... C-74

CSF FAULTS WINDOW..................................................................................................................................... C-76WES 6288 COMPONENTS................................................................................................................................. C-78

WES 6288-1440 DISK ARRAYS ......................................................................................................................... C-80WES 6288-1452 HIGH DENSITY DISK ARRAY ............................................................................................ C-82WES 6289-1440 FIBRE CHANNEL DISK ARRAY......................................................................................... C-84WES 6840-1440/1456 FIBRE CHANNEL DISK ARRAY................................................................................ C-86NCR WORLDMARK ENTERPRISE STORAGE – DESKSIDE.................................................................... C-88

6288 DESKSIDE QUAD MODULAR ARRAY (ONE 1220 ARRAY) ........................................................................... C-886289 FIBER CHANNEL MODULAR ARRAY (ONE 1220 ARRAY)............................................................................ C-88

SMP CONNECTIVITY – SCSI........................................................................................................................... C-90SMP CONNECTIVITY – FIBRE CHANNEL .................................................................................................. C-92MODEL 1440 DISK ARRAY .............................................................................................................................. C-94MODEL 1456 DISK ARRAY .............................................................................................................................. C-96WES MODEL COMPARISON........................................................................................................................... C-98

UNDERSTANDING LSI CONTROLLER NUMBERS.................................................................................................. C-98TW 6.2 SOLUTIONS AND WES 5 PERFORMANCE................................................................................... C-100


Comparing WorldMark SMP ArchitecturesThe SMP is the basic building block of WorldMark Systems. The facing pagecompares the four three basic architectures of WorldMark SMPs.

The NCR WorldMark 5100M (MPP) architecture also supports Teradata Version2 software and uses BYNET V1 technology. The 5100M system was announcedby NCR in 1995 and was available for customer purchase in January 1996. Thisis a previous NCR SMP platform and will not be described in this class. 5100MSMPs utilize the MCA I/O architecture.

The instructor will lead a discussion and provide speeds of the variouscomponents.

Acronyms:MCA – Micro Channel Architecture PCI – Peripheral Component InterconnectEISA – Extended Industry Standard Architecture


Comparing SMP Architectures

4 CPUsPentium Pro

200 MHz

Memory

1 - 4 GB

4700/5150

PCI 0 / Slots - 3 / 4EISA Type - 32 bit / 33 MHzPCI 1 Slots - 3

Type - 32 bit / 33 MHz

System Bus Speed - 66 MHz

4 CPUsPentium Xeon 450 - 700 MHz

Memory

1 - 4 GB

4800/4850/5200/5250 PCI 0 Slots - 2Type - 32 bit / 33 MHz

PCI 1 Slots - 4Type - 32 bit / 33 MHz



4 CPUsPentium Xeon 700 - 900 MHz

Memory

1 - 4 GB

4851/4855/5251/5255 PCI 0 Slots - 2Type - 32 bit / 33 MHz




2 CPUsPentium III

1.4 GHz

Memory

1 - 4 GB

4900/5300 PCI 0 Slots - 2Type - 32 bit / 33 MHz





4700 and 5150 WorldMark Systems The 4700 and 5150 systems are composed of different type of subsystems. Forexample, the basic building block of the system is the “processing node” or SMP.This processing node is implemented in a “chassis” which is mounted in a 4700or 5150 system cabinet. A 4700 or 5150 system cabinet is capable of housing twoprocessing nodes.

The WorldMark 4700 and 5150 systems use industry standard rack mountarchitecture and individual chassis that conform to industry standards. Types ofchassis include:

• Processing Node• BYNET• Server Management• UPS - Uninterruptable Power Supply

The 4700/5150 SMP chassis is 9U in height.


2 SMP Nodes (9U)

4 Intel CPUs Pentium Pro200MHz1 - 4 GB Memory2 PCI Buses

6 PCI slots4 EISA slots

SMP Interconnect Dual BYNET Chassis


Four 3U UPSwith Dual AC


Server Management

Node 1

Node 2

UPS 1

UPS 2

UPS 3

UPS 4

BYNET 1

BYNET 2

4700 and 5150 WorldMark Systems

• 4700 - up to 8 SMPs and uses BYNET 8 switches (V1.1).

• 5150 - up to 128 SMPs and uses BYNET 16/128 switches(V1.1).

• SMPs and components are housed in chassis moduleswhich are mounted in a rack-based cabinet.

• 4700/5150 SMPs use Intel® Pentium® Pro 200 MHz CPUs,a 66 MHz system bus, & leverage 4300 SMP technology.


4800 and 5200 WorldMark SystemsThe 4800 and 5200 systems are similar to the 4700 and 5150 systems in that theyboth utilize the rack-based cabinet. Like other WorldMark systems, the 4800 and5200 systems are composed of different type of subsystems.

One of the key differences is the SMP node – it is based on the Intel 100 MHzinternal bus architecture, uses the Intel Pentium II Xeon 450 MHz or IntelPentium III Xeon 500 or 550 MHz CPUs, and has 3 PCI buses. Simplystated, the 4800/5200 SMP is a faster computing engine than the 4700/5150.

The processing node is housed in an 11U “chassis” which is mounted in a 4800 or5200 rack-based system cabinet. A 4800 or 5200 system cabinet is capable ofhousing two processing nodes.

The WorldMark 4800 and 5200 systems use industry standard rack mountarchitecture and individual chassis that conform to industry standards. Types ofchassis include:

• Processing Node• BYNET• Server Management• Uninterruptable Power Supply


2 SMP Nodes (11U)

4 Intel Xeon CPUs(450/500/550 MHz)

1 - 4 GB Memory 3 PCI Buses 11 PCI slots

Example of BYNET V2(BYNET 16 switch)



4800 and 5200 WorldMark Systems


• 4800 - up to 4 SMPs and uses BYNET 4 switches (V1.1).

• 5200 - up to 512 SMPs and uses BYNET V2 switches.

• 4800/5200 SMPs and components are housed in chassismodules which are mounted in a rack-based cabinet.

• 4800/5200 SMPs use Intel® Pentium® III XeonTM CPUs, a100 MHz system bus, and leverage 4400 SMPtechnology.

UPS 3

UPS 3

UPS 3

SMP - 4 way node

SMP - 4 way node

BYNET V2

BYNET V2

SMC


BYNET V2 Switch Racks (Cabinets)For WorldMark 52xx configurations greater than 16 nodes (SMPs), the BYNETV2 Switch cabinets are needed. There are 3 BYNET V2 Switch cabinetconfigurations as shown on the facing page.

For configurations with 64 or fewer nodes, two BYNET V2 64 Node Switches(BYA64GX) are needed. Each BYNET V2 64 Node Switch is housed in aseparate BYNET V2 Switch cabinet.

For configurations with more than 64 nodes, BYNET V2 Expansion (BYB)Switches are needed. It is also housed in a BYNET V2 Switch cabinet.

The BYNET BYA64GX and BYB64G switches are 12U in height.

The SLAN and PvtLAN switches are located in the back of the cabinet and do notappear as chassis modules on the AWS. These networking components are usedto connect the AWS to the system.


BYNET V2 Switch Racks (Cabinets)

BYNET V264 Node

Switch Rack


Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

BYNET V264/512 NodeExpansion

Rack

BYNET V2Expansion

SwitchChassis

(BYB64G)


Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

BYNET V2512 Node

ExpansionRack

BYNET V2Expansion

SwitchChassis

(BYB64G)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch


Example 1: WorldMark 4855 – 4 NodesNCR offers a four node 4855 system utilizing the BYNET 4 switches.

The 4800 uses the BYA4P switch and BIC2G interface cards. This is a BYNETRelease 1.1 implementation (i.e., 10 MB /sec).

The 4850 uses the BYA4G switch and BIC2C interface cards. This is a BYNETRelease 2.0 implementation (i.e., 60 MB /sec).

The 4855 also uses the BYA4G switch and BIC2C interface cards. This is aBYNET Release 2.0 implementation.

The example on the facing page illustrates a 4855 system with 4 SMP nodes.


Example 1: WorldMark 4855 - 4 Nodes

4855 Dual NodeRack

with BYNET 4switches

4855 Dual NodeExpansion

Rack

SMC (2U)

UPS 2

UPS 1

Disk Array Controllers


SMC (2U)

UPS 2

UPS 1


Disk Array Controllers UPS 1

UPS 2

UPS 3

SMC (3U)

SMP001-4

BIC2C BYA4G

UPS 1

UPS 2

UPS 3

SMC (3U)

6288-1440LSI Disk Array

(WES)


(WES)

Notes:• Each SMP connects to both BYNET 4 switches.• Two 6288 disk array racks can hold up to 160

disk drives - effectively 40 drives for each SMP.

SMP002-4

BIC2C

SMP002-5

BIC2C

SMP001-5

BIC2C BYA4G


Example 2: WorldMark 4900 – 4 NodesNCR offers a four node 4900 system utilizing the BYNET 4 switches.

The example on the facing page illustrates a 4900 system with 4 SMP nodes.



Notes:• Each SMP connects to both BYNET 4 switches.• One 6840 storage rack has up to 112 physical

disks - effectively 28 drives for each SMP.

4900 Rackwith BYNET 4

switches


(WES)

NCR

SMP001-4BIC2M BYA4M

SMC

SMP001-5BIC2M BYA4M

SMP001-6BIC2M

SMP001-7BIC2M

NCR

SMC - 2U

RAID Controllers

RAID Controllers


Example 3: 5300 System - 16 NodesThe WorldMark 5300 system can scale up to 16 processing nodes using theBYNET V2 16 switch (BYA16G). This is a BYNET Release 2 implementation(i.e., 60 MB /sec).


Answer:

4 additional 5300 “Expansion” racks, each with 4 SMPs2 BYNET V2 64 Node Switch racks 4 additional 6840 Storage racks


What additional racks are needed to upgrade thissystem to a 32 node system?


SMCBYNET 16BYNET 16

SMP

SMP

SMP

SMPUPSUPSUPS

SMC

6840-1456

6840-1456UPSUPSUPS

SMC

SMP

SMP

SMP

SMPUPSUPSUPS

SMC

SMP

SMP

SMP

SMPUPSUPSUPS

SMC

SMP

SMP

SMP

SMPUPSUPSUPS

Clique 0

SMC

6840-1456

6840-1456UPSUPSUPS

SMC

6840-1456

6840-1456UPSUPSUPS

SMC

6840-1456

6840-1456UPSUPSUPS

Clique 0


Example 4: 5300 System - 32 NodesThe WorldMark 5300 system can scale to 64 processing nodes using the BYNETV2 64 Node switch (BYA64GX) which is housed in a BYNET V2 Switchcabinet. Note that there are two BYNET V2 Switch cabinets in the illustration onthe facing page.

This is also a BYNET Release 2 implementation (i.e., 60 MB /sec).

What types of cabinets are needed to upgrade this system to 64 nodes?

Answer:

8 additional 5300 “Expansion” racks, each with 4 SMPs8 additional 6840 Storage racks


SMC

6840-1456

6840-1456

SMC

SMPSMPSMPSMP

SMC


BYA64GX

SMC


BYA64GX

What additional racks are needed to upgrade this systemto 64 nodes?


SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

6840-1456

6840-1456

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP

SMC

SMPSMPSMPSMP


Rack (Cabinet) SubsystemsNCR WorldMark MPP systems are built around several key subsystems. Eachsubsystem is implemented in an individual chassis that conforms to industrystandards.

As mentioned before, the types of chassis include:

• Processing or compute node (SMP)• BYNET V2 16 Node switch (BYA16G) – used with 5200/5250/5255

systems with up to 16 nodes• Server Management Chassis (SMC)• Uninterruptable Power Supply – Input Selector (UPS-IS)

The rack/cabinet has been designed to house the different types of chassismodules as shown on the facing page.

Chassis numbers are used to indicate specific components to the AWS.

Chassis numbers are used to indicate specific components to the AWS. The“xxx” in the component names references the cabinet number.


1 CMICxxx-1

2 BYASxxx-2

3 BYASxxx-3

4 SMPxxx-4

5 SMPxxx-5

6 PWRxxx-6

7 PWRxxx-7

8 PWRxxx-8

Rack (Cabinet) Subsystems

Each subsystem is implemented in an individualchassis which conforms to industry standards.

• Processing or compute node (SMP)• BYNET V2 16 Node switch (BYA16G) - optional• Server Management Chassis (SMC)• Uninterruptable Power Supply (UPS)

Chassis AWS 48xx/52xx # Name

1 CMICxxx-1

2 BYASxxx-2

3 BYASxxx-3

4 SMPxxx-4

5 SMPxxx-5

6 SMPxxx-6

7 SMPxxx-7

8 PWRxxx-8 9 PWRxxx-910 PWRxxx-10

Chassis AWS 4900/5300 # Name

BYNET V2 - 16

SMC

UPS 1

UPS 2

UPS 3

4-way SMPBIC4G

BYNET V2 - 16

4-way SMPBIC4G

BYNET V2 - 16 BYNET V2 - 16

SMC

BYNET V2 - 16

BYNET V2 - 16

4-way SMPBIC4M

UPS 3UPS 2UPS 1

4-way SMPBIC4M

4-way SMPBIC4M

4-way SMPBIC4M


4800/4850/5200/5250 SMP Architecture (Logical View)The facing page contains an illustration of major components of the4800/4850/5200/5250 SMP.

This SMP node uses the Intel A450NX (Aspen) Board set components. The mainfeatures of this node are:

• 4800/5200 SMPs may use 4 Intel Pentium II Xeon 450MHz1MB/2MB cache processors (1 MB cache is standard) or 4 IntelPentium III Xeon 500 or 550 MHz CPUs

• 4850/5250 SMPs use 4 Intel Pentium III Xeon 700 MHz CPUs.

• Intel A450NX Intel board Set

• 11 PCI Slots (1 ISA slot is shared with 1 PCI slot)

• 2+1 redundant, Hot Swap power supplies

• 5 Internal SCSI disk drive slots (hot swap). As of June 2000, systemscome standard with four 18GB disk drives.

• 2 SCSI removable bays for CD-ROM and tape

• One 3 ½ ” Diskette drive

• Redundant, front to back cooling (8 fans)

• Server Management – Intel ICMB board and management card (UMB) inSMC

• 11U Chassis


SMP characteristics (a.k.a., Aspen board set)

• 4800/5200 - 4 Intel Pentium II Xeon 450 MHzor Pentium III Xeon 500 or 550 MHz CPUs

• 4850/5250 - 4 Intel Pentium III Xeon 700 MHzCPUs

• 2 GB memory standard, optionally 1 to 4 GB• System bus speed - 100 MHz, 800 MB/sec.• 3 PCI buses, 11 PCI slots, PCI slot #1 is shared

with an ISA slot (not shown)

PCI 32-bit33 MHz Bus #0

System Bus - 100 MHz - 800 MB/sec

Pentium IIIXeon

Pentium IIIXeon

Pentium IIIXeon

Pentium IIIXeon

Memory

4800/4850/5200/5250 SMP Architecture(Logical View)

1 2 3 4 5 6

PCI 32-bit33 MHz Bus #1

7 8 9 10

PCI 64-bit 33 MHz Bus #2

11


4851/4855/5251/5255 SMP Architecture (Logical View)The facing page contains an illustration of major components of the4851/4855/5251/5255 SMP.

This SMP node uses the Intel Koa Board set components. The main features ofthis node are:

• 4851/5251 SMPs use 4 Intel Pentium III Xeon 700 MHz CPUs.

• 4855/5255 SMPs use 4 Intel Pentium III Xeon 900 MHz CPUs. EachCPU has 2 MB of cache. These CPUs are sometimes referred to asCascade processors (internal Intel name).

• 8 PCI Slots (6 full size slots and 2 short slots)


• 5 Internal SCSI disk drive slots (hot swap). As of July 2001, systemscome standard with four 18 GB disk drives.

• IDE based CD-ROM and SCSI tape



• Server Management – Intel ICMB board and management card (UMB) inSMC

• 7U Chassis


SMP characteristics (a.k.a., Koa board set)

• 4851/5251 - 4 Pentium III Xeon 700 MHz CPUs• 4855/5255 - 4 Pentium III Xeon 900 MHz CPUs• 2 GB memory standard

– Physically 1 - 16 GB could be used, butoperating systems limit actual memory

• System bus speed - 100 MHz, 800 MB/sec.• 3 PCI buses, 8 PCI slots (6 full size, 2 short slots)

PCI 32-bit/ 33 MHz Bus #0


Pentium IIIXeon

(700/900)

Pentium IIIXeon

(700/900)

Pentium IIIXeon

(700/900)

Pentium IIIXeon

(700/900)

Memory

4851/4855/5251/5255 SMP Architecture(Logical View)

1 2


5 6 7 8


3 4


4900/5300 SMP Architecture (Logical View)The facing page contains an illustration of major components of the 4900/5300SMP.

This SMP node uses the Intel Dodson Board set components. The main featuresof this node are:

• 4900/5300 SMPs use 2 Intel Pentium III 1.4 GHz CPUs. Each CPU has512 MB of cache. These CPUs are sometimes referred to as Tualatinprocessors (internal Intel name).

• 6 PCI Slots (6 full size slots)


• 5 Internal SCSI disk drive slots (hot swap). As of March 2002, systemscome standard with four 18 GB disk drives.

• IDE based CD-ROM and SCSI tape



• Server Management – Intel ICMB board (IPMI 1.5) and management card(UMB) in SMC

• 5U Chassis


SMP characteristics (a.k.a., Dodson board set)• 2 Pentium III 1.4 GHz CPUs• 2 GB memory standard - 6 interleaved SDRAM

slots– Physically 1 - 6 GB could be used, but

operating systems limit actual memory• System bus speed - 133 MHz, 1066 MB/sec.• 3 PCI buses, 6 PCI slots (6 full size)• Integrated dual redundant 10/100 Ethernet

controllers

4900/5300 SMP Architecture(Logical View)



Pentium III

(1.4 GHz)

Pentium III

(1.4 GHz)


5 6


3 41 2

Memory(6 memory

slots onbaseboard)


MemoryWorldMark 48xx/52xx SMPs use DIMMs (dual inline memory modules) thathave EDO DRAM and utilize four-way memory interleaving for fastest memoryaccess.

WorldMark 4900/5300 SMPs use SDRAM DIMM modules.

4800/4850/5200/5250 MemoryEach 4800, 4850, 5200, and 5250 SMP includes two memory boards bundled asevery unit items. The typical minimum memory for a Teradata node is 2 GB.However, a node can be configured with 1 – 4 GB of memory for Teradata nodes.Physically, these memory boards can support up to 8 GB of memory. However,the MP-RAS operating system is limited to 4 GB.

The following chart shows valid memory configurations. Each bank is actually 4DIMMs. The numbers for memory sizes in this chart represent MB of memory.It is recommended that in multi-node MPP systems, all TPA nodes be configuredwith the same amount of memory.

Memory Board #1 Memory Board #2MemorySize Bank A Bank B Bank C Bank D Bank A Bank B Bank C Bank D1024 10242048 1024 10243048 1024 256 256 1024 256 2563584 1024 256 256 256 1024 256 256 2564096 1024 1024 1024 1024

4851/4855/5251/5255 MemoryEach 4851/4855 or 5251/5255 SMP includes one memory board. The typicalminimum memory for a Teradata node is 2 GB. However, a node can beconfigured with 1 – 4 GB of memory. Physically, this memory board cansupport up to 16 GB of memory. However, the MP-RAS operating system islimited to 4 GB.

4900/5300 MemoryEach 4900/5300 SMP has 6 memory slots on the baseboard. SDRAM DIMMs of512 MB or 1 GB are available. A pair of matching DIMMs must be installed intwo slots. The typical minimum memory for a Teradata node is 2 GB. However,a node can be configured with 1 – 4 GB of memory for Teradata nodes.Physically, the node can support up to 6 GB of memory. However, the MP-RASoperating system is limited to 4 GB.


Memory

• WorldMark SMPs use DIMMs (Dual In-line MemoryModule) and 4-way memory interleaving.

• Teradata (TPA) nodes typically have at least 2 GBof memory. However, SMPs can have 1 to 4 GB.

• Regardless of physical maximums, there aresoftware limitations.– MP-RAS – 4 GB for TPA nodes

– 4 GB for non-TPA nodes– Windows 2000 – 4 GB for TPA nodes

– 8 GB for non-TPA nodes

4800/4850/5200/5250 SMPs

• 2 memory boards – total of 16 DIMM slots.• Physical maximum is 8 GB.

4851/4855/5251/5255 SMPs

• 1 memory board – with 16 DIMM slots.• Physical maximum is 16 GB (NCR limit is 8 GB).

4900/5300 SMPs

• 6 Baseboard DIMM slots – SDRAM.• Physical maximum is 6 GB.


WorldMark PCI I/O AdaptersExamples of PCI adapters used in WorldMark SMPs include:

• PCI Quad SCSI (PQS-2) host adapter - provides access to fourUltra/Wide SCSI buses. Note that this is NOT the same PQS adapter usedwith 4700 and 5150 systems.

• High Performance PQS – provides access to four Ultra 2 SCSI buses(80 MB/sec.) Used with the WorldMark Enterprise Storage – WES(e.g., 6288-1440 Disk Array).

• Quad Fibre Channel (QFC) – provides access to four FC buses (1Gbit/sec.) Used with the WorldMark Enterprise Storage – WES (e.g.,6840-1456 Disk Array).

• BYNET PCI adapters

- BIC2G – used by 4800, connects to BYA4P switch- BIC2C – used by 4850/4851/4855, connects to BYA4G switch- BIC4G – used by 52xx, connects to a BYNET V2 switch- BIC2M – used by 4900, connects to BYA4M switch- BIC4M – used by 5300, connects to a BYNET V2 switch

• PCI Bus ESCON Adapter (PBSA) - provides access to an ESCON fiberchannel. Usually a system will have at least two PBSA adapters forredundancy, even on a small system (e.g., 2 SMPs). This is a newer andfaster channel connection than the PBCA and is the preferred channelconnection adapter.

• PCI Bus Channel Adapter (PBCA) - provides access to an IBMcompatible Bus/Tag (or Parallel) channel connection. The adapterconnects to a Channel Tailgate.

• Networking adapters - there are different PCI network adapters that areavailable to provide connections to different network environments. Forexample, NCR offers PCI Ethernet 10/100 BaseT, GigaBit Ethernet,FDDI, and Token Ring adapters.

Other acronyms:

AUI - Attachment Unit InterfaceBIC – BYNET Interface Card


WorldMark PCI I/O Adapters

Examples of PCI adapters:• PCI Quad SCSI (PQS-2) host adapter - provides

access to 4 Ultra/Wide SCSI buses (40 MB/sec.).

• High Performance PQS host adapter - provides 4Ultra/Wide 2 SCSI buses (80 MB/sec.) - connectsSMPs to WES arrays (e.g., 6288-1440).

• Quad FC (Fibre Channel) host adapter - provides4 Fibre Channel connections (1 Gbit/sec.) -connects SMPs to WES arrays (e.g., 6840-1456).

• BYNET adapters– BIC2G for 4800; connects to the BYA4P switches– BIC2C for 485x; connects to the BYA4G switches– BIC4G for 52xx; connects to BYNET V2 switches– BIC2M for 4900; connects to the BYA4M switches– BIC4M for 5300; connects to BYNET V2 switches

• PCI Bus ESCON Adapter (PBSA) - providesaccess to an ESCON channel.

• PCI Bus Channel Adapter (PBCA) - providesaccess to a Bus/Tag (Parallel) channel.

• Networking adapters - Ethernet 10/100 BaseT,GigaBit Ethernet, FDDI, and Token Ring.


SMP Local Media and DisksEach 48xx or 52xx processing node has local devices (integrated within theprocessing node chassis) to boot the operating system (e.g., UNIX MP-R AS), toinstall or upgrade software, and to backup local files or dumps.

Removable MediaThe node chassis has one 3.5-inch diskette drive and bays for a CD-ROM and atape device.

Disk Drives and SCSI BackplaneThe node chassis has a Ultra/Wide SCSI hot-docking backplane. It supports fivehot-swap bays for SCSI disk drives. These disk are used to hold the UNIXoperating system software, swap space, dump space, and database software.

The disk drives in a 485/5255 are more narrow that the previous 4850/5250. Thebackplane supports five hot-swap bays for 1” wide SCSI drives.

As of June 2001, the default for 48xx/52xx SMPs is that they each have four 18GB disk drives.

UNIX SCSI Device NamesThe typical device names of internal disks for a 4850/5250 SMP are:

- Boot disk /dev/dsk/c100t0d0s0- System disk /dev/dsk/c100t1d0s0- System disk /dev/dsk/c100t3d0s0- System disk /dev/dsk/c100t4d0s0

The typical device names of internal disks for a 4900/5300 SMP are:

- Boot disk /dev/dsk/c200t0d0s0- System disk /dev/dsk/c200t1d0s0- System disk /dev/dsk/c200t2d0s0- System disk /dev/dsk/c200t3d0s0


SMP Local Media and Disks

CD-ROM, SCSI ID 4

BootDisk

0

3.5” Diskette Drive 8mm Tape, SCSI ID 6

SystemDisk

1System

Disk

3Opt.Disk

5

Power Supply

Power Supply

Power Supply

LED

SystemDisk

4

4800485052005250

4851485552515255

49005300

3.5”

Dis

kette

Driv

e

SystemDisk

2BootDisk

0System

Disk

1System

Disk

3Opt.Disk

4

CD-ROM

8mm Tape

3.5” Diskette Drive

CD-ROM

8mm Tape

Opt. Disk

System Disk

System Disk

System Disk

Boot Disk

4

3

2

1

0

11U

7U

5U


Summary of 4855/5255 System ImprovementsThe 4855/5255 Systems have a number of enhancements as highlighted on thefollowing page.

Additionally, the 4855/5255 SMPs also have an improved MTBF – 39,108 hoursversus 35,593 hours for the 4850/5250 SMP.


Summary of 4855/5255 Improvements

SMP Improvements

• Faster 4855/5255 SMPs - approximately 15%faster than 4850/5250 SMPs– 900 MHz CPUs with 2 MB cache– Fast (66 MHz) and Wide (64-bit) PCI bus

• SMP chassis improvements– Smaller chassis - 7U chassis (versus 11U)– Weight is 100 - 110 lbs. (versus 180 lbs.)– Electronics bay within the chassis - higher

level of integration– Improved MTBF

System Improvements

• Server Management enhancements– SLAN uses twisted pair versus coax– SMC has CMIC3 and 10 slots

• faster CPU with more memory whichallows for WEB browser interface

• 10 slots - allows for 4 SMPs in future


Summary of 4900/5300 System ImprovementsThe 4900/5300 Systems have a number of enhancements as highlighted on thefollowing page.

The system will have an improved MTBF (Mean Time Between Failures). Oneof the reasons is that four nodes per cabinet will reduce the number of cabinets(SM Chassis and UPS units, thus improved MTBF).

One 5300 cabinet with 4 nodes has a MTBF of 5,899 hours.

Two 5255 cabinets (each with 2 nodes) has a MTBF of 3798 hours.


Summary of 4900/5300 System Improvements

System Improvements

• Up to 4 SMPs or nodes can reside in a rack -more power in the same footprint.

• Improved system MTBF - four nodes per cabinetusually means fewer cabinets.

• TPA nodes running MP-RAS can be integratedwith non-TPA nodes running Windows 2000 insame system.

• BYNET Release 2.1 - increases the PCI I/Othroughput of the BYNET interface cards (BIC).- New adapters (BIC2M/BIC4M) - support Fast

and Wide PCI buses- New BYA4M switch - supports Fast PCI bus

• Server Management enhancements- uses CMIC3 (introduced with 4855/5255)- utilizes 2U UPS chassis with integrated IS

(Input Selector)

• AWS enhancements- Windows 2000 AWS for MP-RAS systems- AWS can monitor mixed operating systems

in a single MPP system


Summary of 4900/5300 SMP ImprovementsThe 4900/5300 SMPs have a number of performance enhancements as describedon the facing page.

Additionally, the 4900/5300 SMPs also have an improved MTBF – 60,208 hoursversus 39,108 hours for the 4850/4851/5251/5255 SMP (Koa node).


Summary of 4900/5300 SMP Improvements

SMP Improvements

• 4900/5300 SMP– Utilizes Intel Pentium III 1.4 GHz CPUs– Multiple fast (66 MHz) and wide (64-bit) PCI

buses– Simpler SMP design - fewer components

• SMP Node chassis improvements– Smaller chassis - 5U chassis (versus 7U),

allows for 4 SMPs within cabinet/rack– Weight is 70 lbs. (versus 100 lbs.)– Integrated dual 10/100 Ethernet adapters– Improved MTBF– Can be added and/or upgraded in the field


What is BYNET Version 2.1?The facing page lists the performance improvements of BYNET Version 2.1 ascompared to BYNET Version 2.0. BYNET v2.1 is an incremental release toBYNET v2.0 that increases the PCI throughput of the BYNET interface adapters.

The SMP PCI interface (BIC) is changed to 64-bit (Wide) and 66 MHz (Fast). Itis also downward compatible (64 bit or 32 bit @ 66 or 33 MHz). The new BICsare the BIC2M and the BIC4M.

The major benefit is that the PCI I/O throughput of the BIC is quadrupled toeliminate a potential bottleneck. Although it might seem that since a BYNETchannel is full duplex, it could send/receive at 120 MB/sec. In reality, thebandwidth is about 95 MB/sec.

Since there are two BYNETs, then 2 x 95 MB/sec = 190 MB/sec.

However, BYNET v2.0 BICs are PCI 32-bit, 33 MHz adapters and have a PCIthroughput limit of 100 MB/sec. Even though the BYNET can send/receive 195MB/sec., the BIC limits the throughput to 100 MB/sec.

BYNET v2.1 BICs are PCI 64-bit, 66 MHz adapters and have a PCI throughputof 400 MB/sec.

There have been no changes to BYNET v2.0 switch infrastructure with BYNETv2.1. The same BYNET connectors, cables, and switch chassis modules are used.

Another enhancement is the BYA4M switch that supports PCI Narrow/Fast so asto not degrade PCI fast bus.

BYNET v2.1 is interoperable with Bynet v2.0.


What is BYNET Version 2.1?

• An incremental release to BYNET v2.0 thatincreases the PCI throughput of the BYNETinterface adapters.

• The SMP PCI interface is changed to 64-bit (Wide)and 66 MHz (Fast). It is also downwardcompatible (64 bit or 32 bit @ 66 or 33 MHz).– BIC2M for 4900– BIC4M for 5300

• PCI I/O throughput of the BIC is quadrupled toeliminate a potential bottleneck.

• One BYNET channel is full duplex - the actualthroughput is about 95 MB/sec.– Therefore, two BYNET channels is 2 x 95

MB/sec. = 190 MB/sec.– BYNET v2.0 BIC (PCI 32-bit, 33 MHz) PCI

throughput is limited to 100 MB/sec.– BYNET v2.1 BIC (PCI 64-bit, 66 MHz) PCI

throughput is 400 MB/sec.• No changes to BYNET v2.0 switch infrastructure.• BYA4M switch supports PCI Narrow/Fast so as to

not degrade PCI fast bus.• BYNET v2.1 is interoperable with Bynet v2.0.


BYNET Interface Cards (BIC) or Adapters

BIC2G BYNET AdapterThe WorldMark 4800 BYNET uses the BIC2G and BYA4 switches to connect upto 4 SMP nodes. The role of the BIC2G Adapter board is to provide a processingnode with access to two independent BYNET networks via a single PCI slot. OneBIC2G board is required in each SMP node.

The BIC2G is a PCI 2.1 compliant 32 bit / 33 MHz adapter. Following thestandard WorldMark 4800/5200 SMP configuration, the BIC2G adapter isinstalled in PCI slot 1 and is assigned IRQ 5 from the SSU (identified as amultifunction adapter).

The BIC2G adapter (used with 4800 systems) actually has 4 channels or ports.

• Two 125 Megabit/sec. channels, compatible with BYNET V1.0 and V1.1switches.

• Two 1 Gigabit/sec. channels, compatible with BYNET V2.0 switches.

BIC4G BYNET AdapterThe BIC4G adapter board interfaces the WorldMark 4850, 5200, and 5250 SMPnodes to the BYNET network. Each BIC4G provides the circuitry to connect toup to four BYNET Version 2 networks. All four of the ports on the BIC4G arebased on a 1 Gbit/sec Fibre Channel interface.

Important: Only 2 BYNET networks are supported on the current release of theNCR WorldMark systems.

One BIC adapter is housed in each SMP node chassis this adapter provides theinterface to both networks.

The BIC4G is a PCI 2.1 compliant 32 bit / 33 MHz adapter. Following thestandard WorldMark 4850, 5200, 5250 SMP configuration, the BIC4G adapter isinstalled in PCI slot 1 and is assigned IRQ 5 from the SSU (identified as amultifunction adapter).


BYNET Interface Cards (BIC) or Adapters

BIC2G BYNET Adapter

• Used with 4800 SMPs• 4 ports - two V1 channels and two V2 channels

BIC4G BYNET Adapter

• Used with 52xx SMPs• 4 ports - four V2 channels

BIC2C BYNET Adapter

• Used with 485x SMPs• 2 ports - two V2 channels

BIC2M BYNET Adapter

• Used with 4900 SMPs• 2 ports - two V2.1 channels

BIC4M BYNET Adapter

• Used with 5300 SMPs• 4 ports - two V2.1 channels


BYNET SwitchesBYNET 4 Switch (BYA4P) - a PCI card designed to interconnect up to 4 SMPs.This switch is effectively a BYNET V1.1 switch (10 MB/sec.) and is used in the4800. The BYA4P is a PCI card that is placed into a PCI slot of an SMP.

BYNET V2 4 Switch (BYA4G) - PCI card designed to interconnect up to 4SMPs. This switch is a BYNET V2 switch (60 MB/sec.) designed for 485xsystems. The BYA4G is a PCI card that is placed into a PCI slot of an SMP.

BYNET V2.1 4 Switch (BYA4M) - PCI card designed to interconnect up to 4SMPs. This switch is a new BYNET V2.1 switch (60 MB/sec.) designed for 4900systems. The BYA4M is a PCI card that is placed into a PCI slot of an SMP.

BYNET V2 16 Node Switch (BYA16G) – this V2 switch (60 MB/sec.) allows upto 16 5200 SMPs to interconnect. This 3U chassis switch resides in the5200/5300 System Cabinet.

BYNET V2 64 Node Switch (BYA64GX chassis) – this V2 switch is actuallycomposed of 8 BYA8X switch boards in the BYA64GX chassis. Each BYA8Xswitch board allows up to 8 SMPs to interconnect (i.e., 8 switches x 8 SMPs each= 64 SMPs). The BYA64GX is actually a backpanel that allows the 8 BYA8Xswitch boards to interconnect. The BYA64GX also includes a DiagnosticProcessor (DP) board. This 12U chassis resides in either the BYNET V2 64 NodeSwitch cabinet or the BYNET V2 64/512 Node Expansion Cabinet.

Note: BYA8X switch board (in BYA64GX chassis): This is Stage A baseswitch board. Each board supports 8 links to nodes. The BYA64GXchassis can contain a maximum of 8 BYA8X switches, allowing for 64links to nodes. In systems greater than 64 nodes, the BYA8GX switchboards also connect the BYA64GX chassis to BYB64G chassis throughX-port connectors, one on each BYA8X board.

BYNET V2 512 Node Switch (BYB64G chassis) – this V2 switch is actuallycomposed of 4 BYB16G switch boards in the BYB64X chassis. This iseffectively the Stage B expansion switch board. These boards interconnectthrough the BYB backpanel to provide 8 expansion ports. The expansion ports areused to interconnect BYA64GX chassis (through X-port connections on BYA8Xswitch boards). A maximum of 8 BYA64GX chassis can be interconnected to amaximum of 8 BYB64G chassis (one X-port connection from each BYA64chassis to the expansion port on each BYB64 chassis), thus providing up to 512-node switching capacity. This chassis also include a Diagnostic Processor board.


BYNET Switches

BYA4P Switch - Version 1.1 switch

• used with 4800 systems• PCI card that connects up to 4 SMPs

BYA4G Switch - Version 2 switch

• used with 485x systems• PCI card that connects up to 4 SMPs

BYA4M Switch - Version 2.1 switch

• used with 4900 systems• PCI card that connects up to 4 SMPs

BYA16G Switch (BYA16G) - Version 2 switch

• used with 52xx/5300 systems for up to 16 SMPs• 3U chassis that resides in 52xx System Cabinet

BYA64GX Switch - Version 2 switch

• connects up to 64 52xx/5300 SMPs• 12U chassis resides in BYNET V2 Switch Cabinet

BYB64G Expansion Switch - Version 2 switch

• connects up to 8 BYA64GX switches together(8 x 64 nodes = 512 nodes)

• 12U chassis resides in BYNET V2 Switch Cabinet


BYNET Switches – BYA4M and BYA16GThe facing page contains of examples of BYA4M and BYA16G switches.

4700/5150 BYNET Switches (not shown)4700 and 5150 systems used BYNET Version 1.1 switches. These switches arenot shown in this module, but are briefly described in the text.

BYNET V1.1 switches are housed in a 3U chassis that contains the BYNETswitch board (BYA8, BYA16, or BYB32), redundant power supply in a Dual ACconfiguration (1+1), a Chassis Management Board (CMB), and three redundantfans. The BYNET Chassis interfaces to the UPS Subsystem in order to get itspower.

The rear of the BYNET chassis has different types (and number) of connectionsdepending on the type of switch board. All of the switch boards will haveBYNET cable connections, a diagnostic LED, a RS232 connection, an Ethernetconnection, a CMB connection, and power connectors.

The main difference between the BYNET 8, the BYNET 16, and BYNET 128chassis is the switch board in the chassis.

BYNET 8 (BYA8) Chassis - designed to allow up to 8 processing nodes tointerconnect. This will be used for systems that do not have the need forscalability beyond 8 nodes.

BYNET 16 (BYA16) Chassis - designed to allow up to 16 processing nodes tointerconnect.

BYNET 128 (BYB32) Chassis - designed to allow up to 128 processing nodes tointerconnect. The BYB32 switch connects multiple BYA16 switches together.For every 2 BYA16 switches, you need 1 BYB32 switch.

The BYNET chassis for 4700 and 5150 systems is 3U in height. (3 x 1.75” =5.25”)


BYNET Switches - BYA4M and BYA16G

BYNET 0 BYNET 1

SMP4

SMP3

SMP2

SMP1

Where are these BYNET switches located?

Where are these BYNET switches located?

BYNET 0

BYA4M

BYNET 1

BYA4M

. . .SMP2

BYA16G Switch BYA16G Switch

SMP3

SMP1

SMP4

SMP16


BYNET Switches (BYA64GX and BYB64G)The facing page illustrates the purpose of BYA64GX and BYB64G switches.BYA switches connect SMPs. With more than 64 SMPs, multiple BYA switchesare needed for one BYNET and the multiple BYA switches are connectedtogether with BYB switches.


BYNET Switches (BYA64GX and BYB64G)

BYNET 1 is not shown, but all SMPs connect to both BYNETs.

SMP1

SMP2

SMP64

SMP65

SMP66

. . . SMP128

. . .

BYNET 1

BYNET 0BYB switches connectBYA switches together.

SMP1

SMP2

SMP64

. . .

BYA Switch(BYA64GX)

BYNET 0 BYNET 1

SMPs connect to BYAswitches.

BYA Switch(BYA64GX)

BYA Switch(BYA64GX)

BYA Switch(BYA64GX)

BYB Switch(BYB64G)

BYB Switch(BYB64G)


5300 BYNET V2 Example – 128 NodesThe illustration on the facing page shows a 5300 with 128 nodes connecting toBYNET V2 switches.

How many V2 switch cabinets are needed?

Answer: The minimum number if 4. If this system is a new system and initiallyconfigured with 128 nodes, the 4 cabinets are needed.

If this system was initially a 64 node system and then upgraded to 128 nodes, then6 switch cabinets are needed.


5300 BYNET V2 Example - 128 Nodes

How many V2 switch cabinets are needed?

BYNET#1BYA64GX

BYNET#1BYA64GX

BYNET#1 BYB64G

BYNET#1 BYB64G

Each Connection Represents 4Expansion Cables

Node 1 - BIC4M

Node 2 - BIC4M

Node 3 - BIC4M

Node 64 - BIC4M

Node 65 - BIC4M

Node 66 - BIC4M

Node 67 - BIC4M

Node 128 - BIC4M

BYNET#2BYA64GX

BYNET#2BYA64GX

BYNET#2 BYB64G

BYNET#2 BYB64G


BYNET V2 Switch CabinetsThe BYNET V2 64 Node Switch Cabinet can be used for configurations from 2through 64 nodes and must be used for configurations greater than 16 nodes. Allnodes in the configuration are interconnected from the BYNET V2 node interfaceto the BYNET V2 64 Node Switch chassis (BYA64GX). Two BYNET V2 64Node Switch Cabinets are required for the base dual redundant BYNET V2networks.

The BYNET V2 512 Node Expansion Cabinet is for used for configurations thatbegin with 64 nodes or less and has expanded beyond 64 node maximumconfiguration supported by the BYNET BYA64GX chassis (in the V2 64 NodeSwitch Cabinet). Above 64 nodes, the BYNET V2 BYB64G chassis (effectivelya 512 node switch chassis) is used to interconnect multiple BYNET V2 64 nodeswitch chassis. The simple configuration rules are:

• each group of 2 to 64 nodes requires two BYNET V2 64 node switchchassis; a minimum of two is required for dual redundancy.

• for configurations with greater than 64 nodes, each BYNET V2 64 nodeswitch chassis must have a complimentary BYNET V2 512 node switchchassis.

The BYNET V2 64/512 Node Expansion Cabinet is used for configurations thatbegin with greater than 64 nodes. Two of the cabinets provide the BYNET V2switching capability of two BYNET V2 64 Node Switch Cabinet and twoBYNET V2 512 Node Expansion Cabinets.

The SLAN and PvtLAN switches are located in the back of the cabinet and do notappear as chassis modules on the AWS. These networking components are usedto connect the AWS to the system.

The SLAN switch is an 8 port 10BaseT switch and is used to connect the BYNETdiagnostic processors to the SLAN (System LAN).

The PvtLAN switch is a 24 port 10/100 BaseT switch and is used to connectmultiple PvtLAN hubs together for the PvtLAN (Private LAN). The PvtLANswitch is a 2U component.


BYNET V2 Switch Cabinets

BYNET V264 NodeSwitchCabinet


Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

BYNET V264/512 NodeExpansion

Cabinet

BYNET V2Expansion

SwitchChassis

(BYB64G)


Chassis

(BYA64GX)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

BYNET V2512 Node

ExpansionCabinet

BYNET V2Expansion

SwitchChassis

(BYB64G)

UPS 1 2UUPS 2 2U

SMC - CMIC3

PvtLAN Switch

SLAN Switch

The BYNET V2 BYA64GX and BYB64G switches(chassis) reside in BYNET V2 Switch cabinets.


BYNET V2 SwitchesConceptual views of the BYA64GX chassis and the BYB64G chassis are shownon the facing page.


BYNET V2 Switches

Dia

gnos

tic P

roce

ssor BYA8X

BYA64GX Chassis

Dia

gnos

tic P

roce

ssor

BYB

16G

BYB

16G

BYB

16G

BYB

16G

BYB64G Chassis


Example of BYNET V2 Switches - 128 NodesA conceptual view of the BYNET switches for 128 nodes is shown on the facingpage.


Example of BYNET Switches for 128 Nodes

This example only shows one of the two BYNETs.

BYA64GX Chassis

BYB

16G

BYB

16G

BYB

16G

BYB

16G

DP

BYB

16G

BYB

16G

BYB

16G

BYB

16G

DP

BYA64GX Chassis

SMP

BIC4M

BYB64G ChassisBYB64G Chassis

Other BYNET

+ 127 Other Nodes


BYNET™ Logs and UtilitiesBYNET software uses the standard product logging mechanisms to log errors andexceptions.

• AWS Fault Window• /etc/.osm (Console messages)• /var/adm/streams/error.mm-dd (for BYNET messages)• /var/adm/usererr/error.mm-dd (for BIC adapter messages)

In general, there are two methods to diagnose the BYNET and BIC boards:

• Through the SMP based diagnostics software as root user on the SMP (# prompt). • The BYNET utility on the AWS. This utility interfaces with the BYNET

switch board’s Diagnostic Processor. Normally, this utility is used byCustomer (Field) Engineers.

SMP Diagnostic UtilitiesThe following utilities can be used to determine the status of the BYNET from theSMP.

# /usr/bin/blmstat- Shows the status of the SMP’s BLM driver and the BPCI boards.

# /etc/bin/bam -s- (BYNET Administrative Menus) - provides Level 0 diagnostic menus

for the SMP’s BPCI board. These menus are normally used by aCustomer (Field) Engineer. The -s option displays the status of theBYNETs. Options are shown on the facing page.

# /etc/bin/bya4pstat- displays BYA4P or BYA4G statistics


BYNETTM Logs and Utilities

• Standard product logging mechanisms are used:

- AWS Fault Window- /etc/.osm (Console messages)- /var/adm/streams/error.mm-dd (for BYNET)- /var/adm/usererr/error.mm-dd (for BIC)

• In general, there are two diagnostic methods:

- AWS BYNET utility (used by CEs)- SMP utilities (root user - # prompt)

• SMP Diagnostic Utilities

# /etc/bin/bam (BYNET Administrative Menus)

1 / t - run BIC diagnostics2 / s - report BIC to BYNET local connection status3 / r - display connectivity for a single BYNET4 / w - display connectivity discrepancies5 / i - display IP connectivity for all BYNETs6 / d - take a BIC offline7 / u - bring a BIC online8 / e - search streams error log for BYNET messages9 / q - quit


AWS Connectivity to 48xx/52xx SystemTo communicate with processing nodes and system components, the AWS isconnected to the system via two Ethernet LANs.

• SLAN - System Ethernet LAN known as the SLAN; connects AWS to theCMIC in the Server Management Chassis of each cabinet

• PvtLAN – Private Ethernet LAN - connects AWS directly to SMPs.

System LAN (SLAN)The System LAN (SLAN) is a private Ethernet LAN between the AWS andCMICs in a 48xx/52xx and 4900/5300 systems. The MLAN (Management LAN)is an internal LAN that connects components to the CMICs.

• System Events• Console and Diagnostic connections

Private LAN (PvtLAN)This Ethernet LAN is used to directly connect the AWS to each processing node(SMP). This LAN is different from the 4700/5150 in these ways:

• All SMPs are connected using 10baseT (Twisted Pair).

• Up to 100 Mbps full duplex to AWS.

• The PvtLAN uses hubs and switches to allow connection to all nodes.SMP nodes are connected using 8 port 10BaseT hubs (Allied TelesynMR820TR). A 24 port 10/100BaseT switch (Allied Telesyn 8124XL) isused from the AWS to each 8 port hub.

Message Forwarding Not Required - 4700/5150 SMPs that did not have a directPvtLAN connection had their LAN connections and a number of the AWS Statusdisplays routed over the BYNET. Since all 48xx/52xx SMPs have a directPvtLAN connection this message forwarding is no longer required for the48xx/52xx.


AWS Connectivity to 48xx/52xx System

AWS provides the GUI to ... - connect to SMPs - power on/off/reset components - manage faults - obtains h/w or s/w status information

UPS-IS 1

UPS-IS 2

UPS-is 3

SMC

6288-1440Modular

Disk Array

UPS 1

UPS 2

AWS

consolemessages

SLAN

6288-1440Modular

Disk Array

MLANCMIC Collective

CMIC2 - SMC

BYNET 1

BYNET 2

ETHETH

PvtLAN

SMP 2

SMP 1


AWS System Display – Larger ConfigurationThe facing page contains an example of a larger 5200 system configuration.

How many nodes make up this system?

Answer - 28


AWS System Display - Larger Configuration

How many SMP nodes are there in this system?


AWS Menu SelectionsDepending on the system (e.g., 3600, 4700, 4800, 5100, 5150, or 5200) the AWSis connected to, the menu selections may differ slightly. The majority of thefunctions performed by an operator/administrator will be done throughselections/functions that are listed in the pull-down menus.

When you first access the AWS, you will notice that many of the menu functionswill be grayed out. In order to access these functions, you will have to use theAccess menu and you will be prompted with a logon dialog box. In order toaccess these functions, you need either supervisor (root) login privileges on theAWS or a logon name that is associated with one of the AWS administrativegroups.

Some functions in parenthesis (e.g., Cabinet Locator) may be disabled (grayedout) if they are not available to a particular system. For example, the “CabinetLocator” function only applies to a 3600 system.


AWS Menu Selections

CommandConnectConnect to DB WindowExit

EditAdmin Workstation Config

Global ValuesProcessor TypesNode RecordsRescan

System RescanCabinet Rescan

Save MapMark Icons

ManagementPower Control

System Power OnSystem Power OffCabinet Power OnCabinet Power Off(Reset EPO Latch)MarginingCRU/FRU Power OnCRU/FRU Power Off

Subsystem MaintenanceSystem ResetSubsystem ResetReplace CRU/FRUEEPROM FW FlashSEEPROM Rev UpdateMemory DumpBYNET Utility

Configure ProcessorsConfigure AWSConfigure 3600Configure WorldMarkAbort Configuration

AccessEnable All Security FunctionsDisable All Security FunctionsEnable AWS ConfigDisable AWS ConfigEnable Connect to DBDisable Connect to DBEnable Management FunctionsDisable Management FunctionsEnable Console ConnectionsDisable Console Connections

StatusEvent LogUsers Logged OnProcessor StatusJobsFree Disk SpaceInstalled SoftwareKernel ConfigurationBroadcast Msg to UsersDisplay Icon Status ColorsCabinet Status ReportSystem Status Report(Cabinet Locator)Compare Node Software

TestSystem Verification

Node VerificationStart TestDetail Status

MLAN Integrity TestUPS Self TestUPS Input Selector Test

Diagnostic Communications

HelpOn HelpIndexOn Version


Windows 2000 AWS Menu SelectionsThe facing page shows the various menu selections that are available with theWindows 2000 AWS.


Windows 2000 AWS Menu Selections

FileSave Results Panel …Exit

ViewRefresh F5Clear Results PanelSelect Tree Components …Deselect Tree ComponentsUser PreferencesProperties …

FunctionsConnect …FlashPower OnPower OffResetDump Memory

Event MgmtView Events Ctrl + EClear Severity Ctrl + S

ConfigurationSet SCSI IDsSEEPROM UpdateRemove Component DELETE

UtilityBynet UtilityDSSP Main

HelpHelp TopicsAbout Administration Workstation


Command MenuThe command menu contains these menu selections as shown on the facing page.

• Connect• Connect to DB Window• Exit

Connect (to Processor)This function is used to directly connect to a processor (usually an SMP) in thesystem. You must have previously selected the processor(s) that you want toconnect to.

You will be given a choice of three types of connections:

Console (RS-232) specifies a console type connection using the SLAN or theprivate Ethernet LAN. When connecting to an SMP UNIX node in acabinet, this connection is using the CMIC module in the cabinet.

This type of connection is used when a console connection to an SMPnode is required.

LAN specifies an Ethernet LAN connection using the public Ethernet LAN.

You should use this type of connection for normal (most) SMP logins.

Debug specifies a special diagnostic connection to the debug port (a logicalport) of the CMIC. This is a read-only port. By opening this window, asupport engineer can see all of the low-level informational messages thatare generated at the cabinet level, but are not normally passed on to theAWS.

You will be given a choice of two types of window interfaces:

Terminal - A terminal emulation window normally used when connecting toan SMP.

VGA - a formatted window display used with console connections to theDiagnostic Partition of a 4700/4800/4850/5150/5200/5250 SMP. In orderfor the menus to be properly displayed in Diagnostic Partition, use theVGA format. The Diagnostic Partition is accessed by entering “d” duringthe POST tests of these types of SMPs.


Command Menu

Command Edit Management Access Status Test Help

ConnectConnect to DB WindowExit

Connect (to Processor)

Choice of 3 types of connections:

Console (RS-232) - console connection to SMPusing the SLAN.

LAN - specifies an normal LAN connection toSMP; uses the private Ethernet LAN or PvtLAN.

Debug - specifies a special diagnostic connectionto the debug port (a logical port) of the CMIC -uses the SLAN.

Choice of 2 types of window interfaces:

Terminal - an xterm type window.

VGA - used with accessing the DiagnosticPartition.


Connect to an SMP ProcessorThe example on facing page shows the Connect to Processor window. Asshown, the three types of connections are:

Console - for console connections

LAN - for normal Ethernet connections.

Debug - for a read-only diagnostic connection to the debug port. Normallythis option is used with SMPs.

As shown, the two choices for the window interface are:

Terminal - normal terminal emulation window.

VGA - a formatted window display used with console connections to theService/Diagnostic Partition of a 4xxx/5xxx SMP. In order for the menusto be properly displayed in Diagnostic Partition, use the VGA format. TheService/Diagnostic Partition is accessed by entering “d” during the POSTtests of a 4xxx/5xxx SMP.

In this example of connecting to an SMP processor, the selections are Consoleand Terminal.

Once connected to the processor, you may login as any valid user (e.g., root).


Connect to an SMP Processor


Connect to DB WindowIf the Teradata Database is installed on your system, the AWS menu option“Connect to DB Window” can be used to execute the xdbw application.

xdbw is an X Windows application that provides a database (DB) consoleinterface for Teradata. From this interface, a database administrator can executeTeradata utilities and perform database functions such as restart.

The facing page contains an example of the base window provided by xdbw.

Exit - Exiting XconThis option of the Command menu is used to terminate xcon and csfgui programs- effectively closing the AWS Window and the Fault Window. If you logged intothe AWS as awadm, you will also exit the X windows display and be logged outof UNIX.

A confirmation dialog box will be displayed to verify that you really want to exit.

Note:The recommended way to exit the AWS is to use the Command Exitmenu options. If you access the AWS from a remote workstation, using theQuit option from the title bar menu may leave your workstation connected tothe AWS, but you may not be able to access the AWS windows. Therefore,be sure to terminate the xcon and csfgui (fault window) applications usingExit from the Command Menu, rather than just closing the display window.


Connect to DB Window

To start Teradata utilities such as qryconfig,qrysessn, showspace, etc., use the Supvr(Supervisor) function to "start" the application.

Exit - Exiting Xcon

ConnectConnect to DBWindowExit

Command Edit Management Access Status Test Help


AWS Event LogThe AWS maintains an “Event Log” of information that is generated across theSLAN (System LAN). This information comes to the AWS in the form of“event” messages. The event log can be viewed to understand the state of thesystem.

The AWS event log contains “event messages”. With the 3600, the awlogddaemon listened to all of the diagnostic LAN RS-232 ports and places everymessage into the console log.

On CMIC-based systems, the information at the cabinet level is collected by theCMIC which listens to the internal console connections for each processor. TheCMIC (by default) does not send all of the information to the AWS. For example,all of the SUS startup messages are not sent by the CMIC to the AWS. If a SUSerror occurs, it will be sent to the AWS as an event. Examples of information thatis passed to the AWS include:

• Event messages from SMP nodes• SUS Errors• Faults

Circular File The event log is a 512 KB circular file; it contains the last 512 KB of event loginformation. When the event log has 512 KB of old information—informationthat hasn't been saved—the AWS log daemon will copy the file contents to anold_logs file named

/var/console/old_logs/log.date_time

The old logs can be purged periodically by a cron activated script.

TipsThe following scripts can be used from the AWS command line to view the/var/console/console.log.

• /usr/sbin/copy_log - displays the entire console log in chronological order(pipe into more or pg)

• /usr/sbin/tail_log - displays the last page of console log information

When reviewing a fault or other error, you may wish to view the console log tocheck system activity prior to the fault.


AWS Event Log

AWS log of all events transmitted to the AWS on theSLAN and Diagnostic LAN.

awlogd writes these event messages into the file:

/var/console/console.log

The event log has these characteristics:

• 512 KB circular file

• Contains the last 512 KB of information

• When the event log has 512 KB of oldinformation, the contents are copied to a file inthe old_logs directory.

/var/console/old_logs/log.date_time

Typical reasons for viewing the event log:

• View power-up and power-down sequences.

• Review system activity when faults or other errorindicators suggest system problems.


CSF Faults WindowThe CSF Faults Window provides access to outstanding faults and messages inthe system. This window is also used to customize CSF and use the monitoringand tracking features of CSF.

The Fault window is a small, resizable and movable window that is automaticallydisplayed on the AWS. When you click the Faults button of the Fault window,the Fault List window will open and show the currently outstanding faults withone line of information.

From this window, you can select menu commands to perform such activities asviewing faults, deleting faults, reporting problems and reviewing recommendedsolutions.

The Fault List window has a filtering function that allows you to filter outspecified faults. For example, if the filtering option is used, the Fault windowmay contain a value of 12 of 16 indicating that there are 12 faults appearing in theFault List window, but there are really 16 outstanding faults.

To view the fault list from the AWS command line,

# /opt/csf/bin/viewfaults


CSF Faults Window

Fault List

To view the fault list from the AWS command line,

# /opt/csf/bin/viewfaults


WES 6288 ComponentsThe NCR WorldMark Enterprise Storage Solution utilizes a modular arraysubsystem design. The 6288-1440 will have a Controller Module and 4 DiskDrive Modules.

One or two 6288-1440 disk arrays will be housed in the WES cabinet. The modelnumber of the WES cabinet for the WorldMark Enterprise Storage Solution is6000-9122.

Major features include:

• Pentium controllers - Pentium 166 MHz; 32 MB processor memory; 64MB or 256 MB cache memory.

• Point-to-Point SCSI connections to SMPs• Ultra-2 Fast/Wide SCSI buses to SMPs - 80 MB/sec• SMPs requires High Performance PQS adapters


WES 6288 Components

4665 DAC - A4665 DAC - B

Battery

Disk Drive Module(6288-1101)

Controller Module(6288-4665)

Modular Array(6288-1440)

4665 DAC - A4665 DAC - B

Battery

Major features:• Controller module contains dual redundant

Pentium controllers - Pentium 166 MHz; 32 MBprocessor memory; 64 MB or 256 MB cachememory.

• Point-to-Point Ultra-2 Fast/Wide SCSI connectionsto SMPs - 80 MB/sec.

• SMPs requires High Performance PQS adapters.

Example of 6288-1440 (WES 3)


WES 6288-1440 Disk ArraysThe WorldMark Enterprise Storage (WES) Solution uses a rack-based cabinet.One or two 6288-1440 disk arrays are housed in the cabinet.

• Model 6288-1440 - consists of one 6288-4665 controller module and four6288-1101 disk drive modules. The 1440 is 16U tall. This disk array canhave up to 40 disks.

This latest version (July, 2001) of this cabinet has an updated 2U SMC and three2U UPS modules.


WES 6288-1440 Disk Arrays

Features include:• Each 4665 controller connects up to 4 SMPs

using 4 point-to-point SCSI (80MB/sec.) buses.• Dual AC Input and three 2U UPS units (same as

4900/5300).• Service Management - 2U SMC connects to 3U

SMC in SMP cabinet.

NCR

SMC - 2U

4665 Controllers

4665 Controllers

NCR

SMC - 2U

4665 Controllers


WES 6288-1452 High Density Disk ArrayCharacteristics include:

• New drive trays that support a total of 52 low profile hot swappable Ultra-2 SCSI disk drives per array.

• Available disk drives: 18GB 10K rpm LP, 36GB 10K rpm LP, and 36GB15K rpm (March, 2002).

PPlluuss

• Dual hot swappable redundant RAID controllers with either 32MB or256MB of cache.

• SMPs use the High performance PCI Quad SCSI (HP-PQS) Adapter forUltra-2 SCSI point-to-point connectivity (80MB/sec) to multi-ported diskarray.


AAnndd

• Fits into the same WES Cabinet as the 6288-1440.


WES 6288-1452 High Density Disk Array

New drive trays that support a total of 52 low profilehot swappable Ultra-2 Fast/Wide SCSI disk drivesper array.

Characteristics:

• Each drive tray has roomfor 14 drives, but only 13drives per tray areavailable to SCSI.

• Preferred solution for Co-existence solutions of5250 and 5255 systems.

• Available July, 2001 withTW 6.1.

NCR

SMC - 2U

4665Controllers

4665 Controllers


WES 6289-1440 Fibre Channel Disk ArrayCabinet characteristics include:

• New Server Management Chassis.

• New power system - three 2U UPS with Dual AC Distribution Boxes.

• Support for two arrays for a total of 80 drives and 2.0 TB raw datacapacity.


• Dual hot swappable redundant Fibre Channel RAID controllers with either128MB or 256 MB of cache and a AMD 300 MHz K6-2 microprocessor.

• Support for up to 40 hot swappable 1 Gbit Fibre Channel 18GB 10K rpm,18GB 15K rpm, and 36GB 10K rpm disk drives per array.

• Quad Fibre Channel Host Adapter 100 MB/sec (1Gb/sec) point-to-pointconnectivity between processing node and disk arrays.





• Quad Modular Fibre Channel RAID Controllers are designed to match theI/O demands of upcoming 1+ and 2+ GHz nodes.

• Fibre optic point-to-point connections between node and array for higherperformance and greater distances between node and array.


WES 6289-1440 Fibre Channel Disk Array

Available on Windows 2000 with TW 6.1Available on Windows 2000 with TW 6.1

Cabinet characteristics:• New Server Management

Chassis.• New power system - three 2U

UPS with Dual AC DistributionBoxes.

• Support for two arrays for atotal of 80 drives and 2.0 TBraw data capacity.

Fibre Channel Disk Arraycharacteristics:

• Dual hot swappable redundantFibre Channel RAID controllers.

• Support for up to 40 hotswappable 1 Gbit Fibre Channel18GB 15K rpm, and 36GB 10Kand 36GB15K rpm disk drivesper array.

• Quad Fibre Channel HostAdapter (1Gb/sec) point-to-point connectivity betweennode and disk arrays.


NCR

SMC - 2U

4774 Controllers

4774 Controllers


WES 6840-1440/1456 Fibre Channel Disk ArrayCabinet characteristics include:

• New Server Management Chassis.

• New power system - three 2U UPS with Dual AC Distribution Boxes.

• Support for two arrays for a total of 80 drives and 2.0 TB raw datacapacity.


• Dual hot swappable redundant Fibre Channel RAID controllers with either1 GB of cache and an Intel 550 MHz Celeron microprocessor.

• 6840-1440 has support for up to 40 hot swappable 1 Gbit Fibre Channel18GB 15K rpm, 36GB 10K rpm, or 36GB 15K rpm disk drives per array.

• 6840-1456 has support for up to 56 hot swappable 1 Gbit Fibre Channel18GB 15K rpm, 36GB 10K rpm, or 36GB 15K rpm disk drives per array.

• Quad Fibre Channel Host Adapter (1Gb/sec) point-to-point connectivitybetween processing node and disk arrays.





• Quad Modular Fibre Channel RAID Controllers are designed to match theI/O demands of upcoming 1+ and 2+ GHz nodes.

• Fibre optic point-to-point connections between node and array for higherperformance and greater distances between node and array.


WES 6840-1440/1456 Fibre Channel Disk Array

Available on UNIX MP-RAS with TW 6.2Available on UNIX MP-RAS with TW 6.2

Cabinet characteristics:• New Server Management

Chassis.• New power system - three 2U

UPS with Dual AC DistributionBoxes.

• Support for two arrays for atotal of 112 drives and 2.0TBraw data capacity.

Fibre Channel Disk Arraycharacteristics:

• Dual hot swappable redundantFibre Channel RAID controllers.

• Support for up to 40 hotswappable 1 Gbit Fibre Channel18GB 15K rpm, and 36GB 10Kand 36GB15K rpm disk drivesper array.

• Quad Fibre Channel HostAdapter (1Gb/sec) point-to-point connectivity betweennode and disk arrays.


NCR

SMC - 2U

4884 Controllers

4884 Controllers

6840-1456


NCR WorldMark Enterprise Storage – Deskside6288 Deskside Quad Modular Array (one 1220 array)

Characteristics of the 6288 deskside array include:

• Dual hot swappable redundant RAID controllers

• Support for up to 20 hot swappable Ultra-2 SCSI 18GB or 36GB 10K rpmdisk drives.

• High performance PCI Quad SCSI (HP-PQS) Adapter for Wide Ultra-2point-to-point connectivity (80MB/sec) from processing node to multi-ported disk array.

• Support for RAID 1 and RAID 5

• Teradata for MP-RAS Only

6289 Fiber Channel Modular Array (one 1220 array)Characteristics of the 6288 deskside array include:

• One or two fiber channel, hot swappable RAID controllers

• Support for up to 20 hot swappable fiber channel arbitrated loop (FCAL)18GB 15K rpm, or 36GB 10K rpm, or 36 GB 15K rpm disk drives.

• Serial optical fiber channel arbitrated loop (FCAL) interconnect (1Gb/sec)from processing node to multi-ported disk array.

• Support for RAID 1 and RAID 5

• Teradata for Windows 2000


WorldMark Enterprise Storage - Deskside

6288 Deskside Quad SCSI Array (one 1220 array)

• Teradata for MP-RAS Only• Dual hot swappable redundant RAID controllers• Supports up to 20 hot swap 18 GB 15K rpm, 36 GB

10K rpm, or 36 GB 15K rpm disk drives.• SMP uses High Performance PCI Quad SCSI Adapter

for SCSI connectivity (80MB/sec) to disk array• Support for RAID 1 and RAID 5

6289 Fiber Channel Array (one 1220 array)

• Teradata for Windows 2000• One or two fiber channel, hot swappable RAID

controllers• Supports up to 20 hot swap 18 GB 15K rpm, 36 GB

10K rpm, or 36 GB 15K rpm disk drives.• SMP uses QLogic adapter for Fibre Channel

connectivity (1 Gbit/sec.) to disk array• Support for RAID 1 and RAID 5

Deskside Models6288 & 6289


SMP Connectivity – SCSIThe WorldMark Enterprise Storage (WES) 6288 Solution uses a SCSI businterconnect scheme for the Teradata cliques. This scheme uses point-to-pointSCSI bus connections from the host (SMP) to the disk array controller. Thisprovides better performance, higher reliability, and easier cabling andconfigurations.

With the older 6285 disk arrays, a shared SCSI bus and cables were utilized.


SMP Connectivity – SCSI

DAC-A DAC-B

6288-1440

DAC-A DAC-B

6288-1440

DAC-A DAC-B

6288-1440

DAC-A DAC-B

6288-1440

5250SMP

5250SMP

5250SMP

5250SMP

• Each SMP has 2 High Performance PQS adapters• SCSI buses are point-to-point connections• Example of a 4-node clique sharing four WES

6288-1440 disk arrays.


SMP Connectivity – Fibre ChannelThe WorldMark Enterprise Storage (WES) 6840 Solution uses a Fibre ChannelArbitrated Loop (FCAL) interconnect scheme for the Teradata cliques. Thisscheme effectively uses point-to-point Fibre Channel cables from the host (SMP)to the disk array controller. This provides better performance, higher reliability,and easier cabling and configurations.


SMP Connectivity – Fibre Channel

• Each SMP has 2 QLogic Quad Fibre Channeladapters (1 Gbit/sec.)

• Fibre Channel cables are point-to-pointconnections

• Example of a 4-node clique sharing two WES6840-1456 disk arrays.

5300SMP

5300SMP

5300SMP

5300SMP

DAC-A DAC-B

6840-1456

DAC-A DAC-B

6840-1456


Model 1440 Disk ArrayThe 1440 disk array (used in both the 6288 and the 6840 disk arrays) consists ofthe following:

• 1 Disk Array Controller module which contains two 4665 controllers • 4 Disk Drive modules or trays

A minimum of 8 disk drives are required to power up the 1440. Additional diskdrives are available in quantities of 4.

Disks are added from left to right in an array. For example, a 6288-1440 with 8disks would have disks that have SCSI IDs of 0 and 8. The next four disks addedwould have a SCSI ID of 1.

A single disk can be added for the use as a Global Hot Spare. Up to 4 Global HotSpares can be configured for one 1440 array.

The following table shows the 1440 Disk Array capacity.

1440 with 18 GB drives 1440 with 36 GB drives

NumberofDrives

RawCapacity(GB)

RAID 1Capacity(GB)

RAID 5Capacity(GB)

NumberofDrives

RawCapacity(GB)

RAID 1Capacity(GB)

RAID 5Capacity(GB)

8 144 72 108 8 288 144 21612 216 108 162 12 432 216 32416 288 144 216 16 576 288 43220 360 180 270 20 720 360 54024 432 216 324 24 864 432 64828 504 252 378 28 1008 504 75632 576 288 432 32 1152 576 86436 648 324 486 36 1296 648 97240 720 360 540 40 1440 720 1080


Model 1440 Disk Array (SCSI)

1,0 1,8 1,1 1,9 1,2 1,A 1,3 1,B 1,4 1,C

2,0 2,8 2,1 2,9 2,2 2,A 2,3 2,B 2,4 2,C

3,0 3,8 3,1 3,9 3,2 3,A 3,3 3,B 3,4 3,C

4,0 4,8 4,1 4,9 4,2 4,A 4,3 4,B 4,4 4,C

4665 DAC - A4665 DAC - B

Battery

Typically, two sets of mirrored pairs aregrouped together to make up a Vdisk.

MirroredPair in SCSI

Implementation

MirroredPair in SCSI

Implementation


Model 1456 Disk ArrayThe 1456 disk array Fibre Channel Array (used in 6840 disk arrays) consists ofthe following:

• 1 Disk Array Controller module which contains two 6840-4884 controllers • 4 Disk Drive modules or trays

A minimum of 8 disk drives are required to power up the 1456. Additional diskdrives are available in quantities of 4.

Disks are added from left to right in an array. The drives are identified as shownon the facing page.

A single disk can be added for the use as a Global Hot Spare. Up to 4 Global HotSpares can be configured for one 1440 array.

The following table shows the 1456 Disk Array capacity.

1456 with 18 GB drives 1456 with 36 GB drives

NumberofDrives

RawCapacity(GB)

RAID 1Capacity(GB)

RAID 5Capacity(GB)

NumberofDrives

RawCapacity(GB)

RAID 1Capacity(GB)

RAID 5Capacity(GB)

8 144 72 108 8 288 144 21612 216 108 162 12 432 216 32416 288 144 216 16 576 288 43220 360 180 270 20 720 360 54024 432 216 324 24 864 432 64828 504 252 378 28 1008 504 75632 576 288 432 32 1152 576 86436 648 324 486 36 1296 648 97240 720 360 540 40 1440 720 108044 792 396 594 44 1584 792 118848 864 432 648 48 1728 864 129652 936 468 702 52 1872 936 140456 1008 504 756 56 2016 1008 1512

Note: The drive labeling for a 6289-1440 is similar to the 6840-1456. The drivesare numbered 1 to 10 and the drive trays are numbered 1 to 4.


4,1 4,2 4,3 4,4 4,5 4,6 4,7 4,8 … … … … … … 4,14

3,1 3,2 3,3 3,4 3,5 3,6 3,7 3,8 … … … … … … 3,14

2,1 2,2 2,3 2,4 2,5 2,6 2,7 2,8 … … … … … … 2,14

1,1 1,2 1,3 1,4 1,5 1,6 1,7 1,8 … … … … … … 1,14

4884 DAC - A4884 DAC - B

Battery

Model 1456 Disk Array (Fibre Channel)

Typically, two sets of mirrored pairs aregrouped together to make up a Vdisk.

MirroredPair in FC

Implementation

MirroredPair in FC

Implementation


WES Model ComparisonThe chart on the facing page lists the major differences between the WES 6288and WES 6840 Storage Solutions.

Understanding LSI Controller NumbersThe Disk Array Controllers have the following numbering scheme:

abcd

a – Series of Controller3 – Series 34 – Series 45 – Series 5 (future)

b – Controller to Host Interface6 – SCSI7 – 1 Gbit Fibre Channel8 – 2 Gbit Fibre Channel

c – Controller to Drive/ESM Interface6 – SCSI7 – 1 Gbit Fibre Channel8 – 2 Gbit Fibre Channel

d – Number of internal drive channels4 – four internal drive buses or channels5 – five internal drive buses or channels

For example:

4665 ControllerSeries 4, SCSI to Host, SCSI to Drives, and 5 internal drive channels

4884 ControllerSeries 4, 2 Gbit FC to Host, 2 Gbit FC to Drives, and 4 internal drivechannels

Note: Even though the 4884 controller itself is capable of 2 Gbit FCthroughput, the host adapters and disk drives are currently onlycapable of 1 Gbit transfers.


WES Model Comparison

WES 3(6288 Quad Array)

WES 5(6840 Fibre Channel)

ModelsSupported

6288-1220 Deskside6288-14406288-1452

6840-14406840-1456

MaximumDisks perArray/Cab.

2040/8052/104

40/8056/112

HostAdapter

High Performance PQSSCSI QLogic 2204

HostInterface toController

Point-to-PointUltra-2 SCSI(80 MB/sec.)

Fibre ChannelArbitrated Loop

(1 Gbit/sec.)

Controller 4665 Quad SCSI 4884 Quad FC(2 Gbit/Sec.)

ControllerCPU

IntelPentium 166 MHz

IntelCeleron 550 MHz

ControllerCache

32 – 256 MB 1 GB

Controller toDiskInterface

Ultra-2 SCSI(80 MB/Sec.)

Fibre Channel(1 Gbit/sec.)

DrivesSupported

18 GB 10K rpm36 GB 10K rpm36 GB 15K rpm

18 GB 15K rpm36 GB 10K rpm36 GB 15K rpm

LUNsSupported

32 32

RAID Mgt.Software

Raid Manager 5and RDAC

SYMplicity 7.10and MPPD

RAID LevelsSupported

1, 3, 5(Only 1 and 5 are used)

1, 3, 5(Only 1 and 5 are used)


TW 6.2 Solutions and WES 5 PerformanceThe facing page contains a matrix that identifies various storage solutions forNCR WorldMark servers.

Matrix Notes:

1. 6288-1452 is a preferred solution for MPP Co-Existence2. 6840-1440/1456 will be available March, 2002.

Cable Notes:

WES Cabinets- SCSI cable lengths will be 8 and 12 meters.- Fibre cables lengths will be 12 and 100 meters.

EMC Symmetrix 5 Cabinet- SCSI length will be 12 meters.- Fibre Channel cable lengths available will be 10, 50 and 100 meters.


TW 6.2 Solutions and WES 5 Performance

WES 5 Performance

WES 5 raw performance is 92% faster than WES 3.

– WES 3 I/O max throughput = 130 MB/sec.

– WES 5 I/O max throughput = 250 MB/sec.

Note: This matrix is NOT a complete list of possible solutions.

SMP MPP

Windows 6289 Deskside 6289-1440

MP-RAS 6288 Deskside 6288-14406288-14521

6840-14402

6840-14562

1. Preferred for MPP Co-Existence2. Available March, 2002

Examples of TW 6.2 Solutions


Notes