Hadoop and the Data Warehouse: Point/Counter Point
-
Upload
inside-analysis -
Category
Technology
-
view
280 -
download
6
description
Transcript of Hadoop and the Data Warehouse: Point/Counter Point
Grab some coffee and enjoy the pre-show banter before the top of the hour!
The Briefing Room
Hadoop and the Data Warehouse: Point/Counter Point
Twitter Tag: #briefr
The Briefing Room
! Reveal the essential characteristics of enterprise software, good and bad
! Provide a forum for detailed analysis of today’s innovative technologies
! Give vendors a chance to explain their product to savvy analysts
! Allow audience members to pose serious questions... and get answers!
Mission
Twitter Tag: #briefr
The Briefing Room
Topics
This Month: BIG DATA
May: DATABASE
June: ANALYTICS & MACHINE LEARNING
2014 Editorial Calendar at www.insideanalysis.com/webcasts/the-briefing-room
Twitter Tag: #briefr
The Briefing Room
Twitter Tag: #briefr
The Briefing Room
Analyst: Robin Bloor
Robin Bloor is Chief Analyst at The Bloor Group
[email protected] @robinbloor
Twitter Tag: #briefr
The Briefing Room
Teradata
! Teradata is known for its analytics data solutions with a focus on integrated data warehousing, big data analytics and business applications
! It offers a broad suite of technology platforms and solutions and a wide range of data management applications
! Teradata’s SQL-H allows users and applications to join Hadoop data to the Teradata Data Warehouse and the Aster Discovery Platform
Twitter Tag: #briefr
The Briefing Room
Guest: Dan Graham
Dan Graham is the Technical Marketing Director for Teradata. With over 30 years in IT, Dan joined Teradata Corporation in 1989 where he was the senior product manager for the DBC/1012 parallel database computer. He then joined IBM where he wrote product plans and launched the RS/6000 SP parallel server. He then became Strategy Executive for IBM’s Global Business Intelligence Solutions. As Enterprise Systems General Manager at Teradata, Dan was responsible for strategy, go-to-market success, and competitive differentiation for the Active Enterprise Data Warehouse platform. He currently leads Teradata’s technical marketing activities.
Point, Counterpoint Myths and Magic
HADOOP AND THE DATA WAREHOUSE
11 Copyright Teradata
• Words and modern terminology
• Is Hadoop a data integration product?
• Is Hadoop a data warehouse?
• Hadoop – the magic
• Teradata and Hadoop
AGENDA
http://www.teradata.com/analyst-reports/Hadoop-and-the-Data-Warehouse-Competitive-or-Complementary/
12 Copyright Teradata
Our Host, the Word Smith
@RobinBloor
13 Copyright Teradata
14 Copyright Teradata
• At least most of the time
• Queries are significantly faster but not always instantaneous > Simple selects à A couple of seconds > Join queries à 10s of seconds
Now we have something that can provide us “Real-time” in Hadoop
Source: Slideshare, Real Time Interactive Queries IN HADOOP: Big Data Warehousing Meetup, June 2013
15 Copyright Teradata
Term Hadoop meaning BI/DW meaning
Real time query
• Self- service interactive queries that run in under minutes, preferably < 10s of seconds
• Query responses in milliseconds • +Advanced query prioritization
SQL • Subset of ANSI 92 SQL • Primary data types • UDFs
• ANSI 2008 SQL + • Some/all ANSI SQL 2011 • All SQL data types • Integrity constraints, window
functions, UDFs, triggers, XML • ACID transactions (start
transaction, commit, rollback) • Geospatial, temporal
OLAP • Any query < 10 seconds
• Subsecond multi-dimensional aggregate queries
• Roll-up, drill-down hierarchies • MOLAP and ROLAP
Hadoop Translator
See: Wikipedia
16 Copyright Teradata
q Real-Time Query: Real-Time is really business time. It is almost always performance critical (otherwise why would you engineer for it?).
q SQL sophistication depends on what you want to use it for. SQL-92 is rather primitive. There are consequences – performance consequences.
q The appropriateness of Hadoop Interactive (OLAP) capability is user dependent. But why would you use Hadoop for this?
Shoop, Shoop Hadoop!
17 Copyright Teradata
Current HDFS Availability & Data Integrity
• Simple design, storage fault tolerance > Storage: Rely in OS’s file system rather than use raw disk > Storage Fault Tolerance: multiple replicas, active monitoring > Single NameNode Master – Persistent state: multiple copies + checkpoints – Restart on failure
• How well did it work? > Lost 19 out of 329 Million blocks on 10 clusters with 20K
nodes in 2009 – 7-9’s of reliability – Fixed in 20 and 21.
> 18 months Study: 22 failures on 25 clusters - 0.58 failures per year per cluster – Only 8 would have benefitted from HA failover!! (0.23 failures
per cluster year) > NN is very robust and can take a lot of abuse – NN is resilient against overload caused by misbehaving apps
Source: Slideshare, NameNode HA, 2011
18 Copyright Teradata
Term Hadoop meaning
BI/DW meaning
High Availability
• Data replication • Name node fail
over
• Redundant access paths (network, nodes, disks)
• RAID storage, high quality hardware • Minimized planned downtime • No single point of failure • HA administration tools, event alerts
tracking and auto recovery • Backups
Fault tolerant
Query automatically restarts on another node without resubmission using replicated data
• Nonstop system (no unplanned system halt or reboot)
• Extreme hardware reliability • 99.999% uptime • Fault isolation and containment • Graceful degradation • Rolling upgrades
Hadoop Translator
19 Copyright Teradata
Hadoop Falling Over!
q Hadoop was built for the recovery of large batch on large commodity grids. q The goal was not to lose the work q This is really about disk failure
q HA/FT is always configured according to workload characteristics. Enterprise HA is best thought of as “transactional” and OLTP, at the least, if not a real-time event.
20 Copyright Teradata
Is Hadoop a Data Integration Platform?
• Yes > “Lots of customers doing
ETL in Hadoop” > Data refineries > Unstructured data – Weblogs and sensor data
> Data Hub/Data Lake
• No > No built-ins – Data quality tools – Transformations
> All do-it-yourself code > No ETL process
management > No metadata repository
21 Copyright Teradata
• Data integration requires a method for rationalizing inconsistent semantics, which helps developers rationalize various sources of data (depending on some of the metadata and policy capabilities that are entirely absent from the Hadoop stack).
• Data quality is a key component of any appropriately governed data integration project. The Hadoop stack offers no support for this, other than the individual programmer's code, one data element at a time, or one program at a time.
• Because Hadoop work streams are independent — and separately programmed for specific use cases — there is no method for relating one to another, nor for identifying or reconciling underlying semantic differences.
Hadoop Is Not a Data Integration Solution
29 January 2013
22 Copyright Teradata
• Facebook, Twitter, LinkedIn • Sensor data • XML • Web logs • JSON • eMail • Documents • Images
• Not so much > Audio > Video
Unstructured Data in the Data Warehouse
20%
15%
10%
5%
0% Social Docs eMail JPGs A/V Sensor XML Web
logs
2013
Sources: Derived from TDWI, Wikibon, Gartner, IDC
23 Copyright Teradata
Hadoop For Data Integration!
q Hadoop serves a useful function as a data reservoir. q The revenge of the ISAM file q Some ETL q Some cleansing q Some analytics
q Personally, I would want drag and drop ETL, ELT. Those who write code maintain code.
24 Copyright Teradata
Is Hadoop a Data Warehouse?
25 Copyright Teradata
At Facebook, we have unique storage scalability challenges when it comes to our data warehouse. Our warehouse stores upwards of 300 PB of Hive data, with an incoming daily rate of about 600 TB. In the last year, the warehouse has seen a 3x growth in the amount of data stored. Given this growth trajectory, storage efficiency is and will continue to be a focus for our warehouse infrastructure. There are many areas we are innovating in to improve storage efficiency for the warehouse – building cold storage data centers, adopting techniques like RAID in HDFS to reduce replication ratios (while maintaining high availability), and using compression for data reduction before it’s written to HDFS. The most widely used system at Facebook for large data transformations on raw logs is Hive, a query engine based on Corona Map Reduce used for processing and creating large tables in our data warehouse. In this post, we will focus primarily on how we evolved the Hive storage format to compress raw data as efficiently as possible into the on-disk data format.
Scaling the Facebook Data Warehouse to 300 PB
April 10, 2014 https://code.facebook.com/posts/229861827208629/scaling-the-facebook-data-warehouse-to-300-pb/
26 Copyright Teradata
• A data design pattern, an architecture > Size doesn’t matter > A perpetual evolution
• Definition: Gartner (2005) /Inmon (1992) /Wikipedia > Subject oriented – Detailed data + modeling of sales, inventory, finance, etc.
> Integrated logical model – Merged data – Consistent, standardized data formats and values
> Nonvolatile – Data stored unmodified for long periods of time
> Time variant – Record versioning or temporal services
> Persistent storage, not virtual, not federated
What is a Data Warehouse?
Source: Gartner: Of Data Warehouses, Operational Data Stores, Data Marts and Data 'Outhouses‘; Bill Inmon, Building the Data Warehouse, 1992, Wiley and Sons
27 Copyright Teradata
Subject Areas: A Model of ‘Our’ Business
Price history
Inventory Supplier
Contracts
Product/Services
Channels
E-Commerce
Labor
Associate
Customer
Sales transactions
Point of Sale
Shipment Carrier
Campaigns
Promotion
Warehouse
Each subject area has numerous large FACT tables (=big joins)
28 Copyright Teradata
You Wish You Had Redundant Data!
App Cust_ID First Last DOB Social Address
ERP 30391-244 William Franks 04/12/00 563-49-1234 123 Oak, Atlanta
CRM 30391244 W. Franks 04/12/70 563491234
SCM 30391244 Bill Franks 04/12/70 Atlanta
XYZ 30391-244 Frank Williams 563491234 123 Oak St. #14
Cust_ID First Last DOB Social Address
30391244 William Franks 04/12/70 563491234 123 Oak St. #14
Final integrated record
Match keys
ETL
29 Copyright Teradata
• A targeted project that will be finished > A subset of data, not all the data > Not for all of the people
• Often heavily denormalized
• Volatility > Often completely reloaded
• Time variance and currency > Can restate the data “as of” a point in time
• Virtualization option > Can be a logical set of views, cubes
What is a Data Mart?
Source: Gartner: Of Data Warehouses, Operational Data Stores, Data Marts and Data 'Outhouses‘; Inmon, Building the Data Warehouse, 1992, Wiley and Sons
30 Copyright Teradata
Why Hadoop Is Not a Data Warehouse
31 Copyright Teradata
Words Matter!
q The meaning of data warehouse is changing: q JSON (hierarchical capability) q Network queries (possibly offload) q Analytics
q The meaning of data warehouse is extending. But it still includes “optimization.”
q It’s no longer a data staging area, it’s a reservoir.
Where Hadoop Excels
HADOOP MAGIC
33 Copyright Teradata
• Apache Hadoop framework > Hadoop Common – Libraries and utilities
> Hadoop Distributed File System (HDFS) > Hadoop YARN > Hadoop MapReduce
• Looks like Hadoop too > Ambari, Avro, Cassandra, Chukwa, HBase, Hive, Mahout, Pig,
ZooKeeper, Oozie, Sqoop, Tez
What is Hadoop?
• SQL on Hadoop o Presto, Drill, Shark, Impala, Hawq, JAQL
• New Hadoop modules or replacements o GPFS, Mesos, Spark, Storm, Accumulo, Sentry, Falcon,
Knox, Whirr , Sentry, Tachyon, SOLR, Lucene
34 Copyright Teradata
Hadoop 2.0
Applica'ons Run Na'vely IN Hadoop
HDFS2 (Redundant, Reliable Storage)
YARN (Cluster Resource Management)
BATCH (MapReduce)
INTERACTIVE (Tez)
STREAMING (Storm)
GRAPH (Giraph)
IN-‐MEMORY (Spark)
HPC MPI (OpenMPI)
ONLINE (HBase)
OTHER (ex. Search)
Hadoop 2.0
35 Copyright Teradata
• Rapid ingest > File copy vs database load
• Temporary data
• Data not ready for the data warehouse
• Data that never à data warehouse
• Archives > Alternative to magnetic tape
Data Lake Benefits: The Landing Zone
36 Copyright Teradata
• Ad hoc projects > One-shot complex analytics > Hurry up, short term efforts
• Alternative analytics > Not SQL-friendly algorithms > Markov chains, random forest > JPG, audio analysis
• Sandbox – hunting in the dark > Prototyping > Data exploration > Trial and error new algorithms
Hadoop Enables Another Data Platform
37 Copyright Teradata
What Hadoop Is For!
q Data reservoir
q Prototyping
q Analytical or BI sandboxing (data wrangling)
q Archive
q File system API (HDFS)
Math and Stats
Data Mining
Business Intelligence
Applications
Languages
Marketing
ANALYTIC TOOLS & APPS
USERS
INTEGRATED DISCOVERY PLATFORM
INTEGRATED DATA WAREHOUSE
ERP
SCM
CRM
Images
Audio and Video
Machine Logs
Text
Web and Social
SOURCES
DATA PLATFORM
ACCESS MANAGE MOVE
TERADATA UNIFIED DATA ARCHITECTURE System Conceptual View
Marketing Executives
Operational Systems
Frontline Workers
Customers Partners
Engineers
Data Scientists
Business Analysts
TERADATA DATABASE
HORTONWORKS
TERADATA DATABASE
TERADATA ASTER DATABASE
39 Copyright Teradata
• Joint R&D with Hortonworks > Donated to Apache
• Business user query with favorite BI tools
• Join Hadoop data to > Teradata Data Warehouse > Aster Discovery Platform
• Teradata 15.0 > Bi-directional SQL > Push down filters to Hive
• Fast, secure, reliable
Teradata SQL-H Teradata SQL-H Aster SQL-H
Hadoop Layer: HDFS
Pig
Hive
Hadoop MR
HCatalog
Dat
a
Dat
a Fi
lter
ing
40 Copyright Teradata
Teradata 15: Teradata QueryGrid™
SQL, SQL-MR, SQL-GR
TERADATA ASTER
DATABASE
Teradata Systems
TERADATA DATABASE
OTHER DATABASES
Remote Data
LANGUAGES
SAS, Perl, Python, R, Ruby, etc,
HADOOP
Push-down to Hadoop
IDW Discovery
TERADATA DATABASE
TERADATA ASTER
DATABASE
Business users Data Scientists
41 Copyright Teradata
Market Possibilities
q The scale-out file system will not die (because it’s only an API)
q YARN (& Cascading) will prosper
q Hadoop will play a role in data flow
q It will never replace the EDW, except by deception
q The struggle for a unified architecture will continue
42 Copyright Teradata
Hadoop and the Data Warehouse: Competitive or Complementary?
http://www.teradata.com/analyst-reports/Hadoop-and-the-Data-Warehouse-Competitive-or-Complementary/
Twitter Tag: #briefr
The Briefing Room
Twitter Tag: #briefr
The Briefing Room
Upcoming Topics
www.insideanalysis.com
2014 Editorial Calendar at www.insideanalysis.com/webcasts/the-briefing-room
This Month: BIG DATA
May: DATABASE
June: ANALYTICS & MACHINE LEARNING
Twitter Tag: #briefr
The Briefing Room
THANK YOU for your
ATTENTION!