Database System Architectures

Client-server Database System Parallel Database System Distributed Database System

Wei Jiang

Centralized Systems

CPUDisk

ControllerPrinter

ControllerTape-driveController

Memory Controller

Memory

System bus

Databases are designed in two ways due to Computer SystemSingle-user system• A desktop unit used by a single

person

• Only one CPU, one or two disk

Multi-user System• It serves a large number of users who

are connected to system via terminals• More disks, more memory, multiple

CPUs

Database

Designed by Single user•Not support concurrency control•Provision for crash-recovery are either absent or primitive•Do not support SQL,but provide a simpler language, such as QBEDesigned by multi-userSupport full transactional features

Currently development

Databases designed for Single-processor machines provide multitasking

Client-Server Systems

General structure of a client-server SystemClient Client Client Client

Server

….

Front-end and back-end functionality

SQL uer-interface

Formsinterface

Reportwriter

Graphical interface

SQL engine

Front-end

Interface

(SQL +API)

Back-end

Server System Architectures

Transaction-server Systems

Provide an interface to which client can send requests to perform an action, in response to which they execute the action and send back results to the client. Requests may be specified by using SQL, or through a specialized application program interface

Data-server SystemsAllow clients to interact with

the servers by making requests to read or update data, in units such as files or pages.

For example, file servers provide a file-system interface where clients can create, update,read and delete files.

Transaction Server Process Structure• Server Processes: receive user queries, execute them and sent result back

• Lock manager process: implements lock manager functionality, which includes lock grant, lock release, and deadlock detection

• Database writer process: for output modified buffer blocks back to disk

• Log writer process: process output log records from the log record buffer to stable storage

• Checkpoint process: performs periodic checkpoints

• Process monitor process: monitors other processes, and take recovery actions for the failed process

The shared memory:

Buffer pool, Lock table, Log buffer,Cached query plans

Mutual exclusion rule

Data Servers• Prefetching page shipping: when an item (both tables and tuples) is

requested, it also sends back other items that are likely to be used in near future

• De-escalation Locking: if the client does not need a prefetched item, it can transfer locks on the items back to the server, and the lock can be allocated to other clients

• Data Caching Coherency: even if a transaction finds cached data, it must make sure that those data are up to date. A message must still be exchanged with the server to check validity of the data, and acquire a lock on data.

• Call back lock caching: If a client requests a lock from the server, the server must call back all conflicting locks on the data item from any other client that have cached the locks.

Parallel Systems• Purpose: improve processing and I/O speeds

• Two kinds: 1)a coarse-grain parallel machine 2)a massively parallel machine

• Measures of performance: 1)throughput: the number of tasks that can be completed in a time interval

2)response time: the amount of time it takes to complete a single task from the time it is submitted

• Two important issues: speedup and scaleup

SpeedupDef: Running a given task in less

time by increasing the degree of parallelism

Goal: to process the task in time inversely proportional to the number of processors and disks allocated.

Calculation: speedup is Ts/Tn when the larger system has N times the resources of smaller system

Type: 1)linear speedup: N 2)sublinear speedup: <N

speed

resources

Linear speedup

Sublinear

speedup

Scaleup Def: the ability to process

larger tasks in the same amount of time by providing more resources.

Calculation: Ts / Ta (Ts-execution time of tasks Q on

machine with size Ms;

Ta-execution time of tasks N*Q on machine with size N*Ms)

Type: 1)linear scaleup: Ts = Ta

2)sublinear scaleup: Ta > Ts

Linear scaleup

Sublinear scaleup

Problem size

Ts/Ta

Scaleup ContinueTwo kinds of scaleup Batch scaleup: The large jobs whose runtime depends on the

size of database. Transaction scaleup: The rate at which transactions are

submitted to the database depends on the size of the database .

The factors work against efficient parallel operation Startup costs:the startup time overshadow the actual processing

time, affecting speedup adversely Interference: it competes with existing processes for commonly

held resources Skew: it is often difficult to divide a task into exactly equal-

sized parts

Interconnection Networks

Bus: send and receive data from a a single communication bus.It can handle communication from only one component at a time. Work well for small numbers of processors, but not scale well with increasing parallelism

Mesh: The number of communication links grows as the number of components grows, and work well with increasing parallelism

Hypercube: The components are numbered in binary, each of the n components is connected to log(n) other components. The communication delays are lower than mesh

bus mesh hypercube

Parallel Database ArchitecturesShared Memory:

--Def: The processors and disks have access to a common memory via a bus or through an interconnection network.

--strength: extremely efficient communications between processes

--weakness: Maintaining cache increasing overhead

Shared Disk:

--Def: all processors can access all disks directly, but the processors have private memory.

--strength: 1) each processor has its own memory, the memory bus is not a bottleneck.

2)if a processor fails, the other processors can take over its tasks.

--weakness: The interconnection to the disk subsystem is now a bottleneck.

Parallel Database continued

Shared Nothing:

--Def: each node of machine consists of a processor, memory, and one or more disks.

--strength: more scalable and easily support a large number of processors.

--weakness: cost high

Hierarchical:

--Def: combines the above three

Distributed SystemDef: The database is stored on several computers which communicate with each other through communication media.

The main difference between Distributed database and share-nothing parallel database:

1)the databases are separated geographically, so a slow interconnection

2)there exist a local transaction and global transaction

The reasons for building Distributed System:

1)sharing data

2)Autonomy

3)Availability

The end