Distributed, Parallel and Internet Databases

1
Distributed, Parallel and Internet Databases

• David Nelson
• CAT
• January 2004

2
Contents

• Definitions
• Distributed Database Systems
• Homogeneous and Heterogeneous Systems
• Federated DBMS Systems
• X/OPEN DTP Standard
• Internet Databases
• Client/Server Architectures, advantages and
  disadvantages
• Web Database Approaches
• Further Reading

3
Definitions

• Distributed Database System
• the ability of the DDBS users to run applications
  at each node
• Parallel Database System
• black box nodes
• Federated Database System
• a DDBS is usually a single application
  distributed over various sites
• a FDBS is a cooperating multiple system

4
Traditional Architecture

• Traditional Database Systems are based on a
  two-tier client-server architecture

• User interface
• Main business and data processing logic

Client
Database Server

• Server-side validation
• Database access

5
Distributed Database Systems

• System needs facilities to be able to
• perform distributed query optimization
• manage distributed transactions
• manage data replication
• Homogeneous DDBS simplest case
• several sites, each running applications on same
  DBMS with same schema and transactions
• location transparency
• can communicate over large distances, and are
  autonomous

6
Heterogeneous DDBMS

• Several existing databases (using different
  DBMSs) linked into a single system
• Problems
• variation in costs of operation between sites
• some operations may not be available at some
  sites
• some DBMSs cannot read records of others
• varying base types
• Requesting site must
• have detailed knowledge of operation of remote
  system
• assume remote system has only rudimentary
  functionality
• make programmer do query composition by hand

7
Federated DBMS

• A collection of independently managed,
  heterogeneous database systems
• allow partial and controlled sharing of data
  without affecting existing applications

Federated schema
Federated to local schema mapping
Local schema
Federated schema
Federated to local schema mapping
Local schema
8
Client-Server Systems

• DB applications can be classified according to
  granularity of database accesses
• coarse grained - transactions which access large
  volumes of data, e.g. decision support, large
  scientific apps
• medium grained - access groups of up to several
  dozens of records, e.g. order entry, convential
  commercial apps)
• fine grained - typically navigational, accessing
  small portions of linked records, e.g. CAD, CASE
  etc.

9
Coarse Grained Systems

• Few large interactions with database
• will benefit from parallel execution of queries
  on the server
• may be necessary to bring back useful data to
  client for sequential processing

Application (on client)
Application (on client)
Database server
10
Fine Grained Systems

• Often advantageous to request large data
  transfers from server to local cache
• high locality of reference gives good performance
• built-in inter-query parallelism

application client database (cache)
application client database (cache)
Database server
11
Data Placement

• Two aims in spreading the data over the nodes
• balance load between nodes (or disks)
• minimize data transfers between nodes
• Aims are incompatible - need a compromise
• place frequently co-referenced records on the
  same node
• ensure groups and evenly distributed between
  nodes
• Optimal size depends on range of cost-effective
  granule sizes, and workload characteristics

12
Parallel Database Architectures

• Classified according to the elements that are
  shared
• Shared Memory Systems
• Shared Disk Systems
• Shared Nothing Systems
• Hybrid Systems

13
Shared Memory Systems

• All processors share disks and main memory
• Inter-query parallelism achieved by adding more
  processors
• intra-query difficult
• Does not scale well beyond 20 processors
• Many commercial systems (Oracle, DB2, Ingres)

...
processor
processor
processor
Memory
...
14
Shared Disk Systems

• Consists of nodes (processors and memory) and a
  pool of shared disks
• Easy to port existing systems
• by mapping transactions to single nodes (gives
  inter-query parallelism)
• Limited by volume of traffic on communications
  network

...
Processor memory
Processor memory
Processor memory
...
15
Shared Nothing Systems

• Nodes linked by a communications network
• Data allocated to nodes
• every query must be parallelised
• load balancing is difficult, as data is
  statically distributed
• Heavily reliant on intra-query parallelism
• e.g. ICL Goldrush

...
Processor memory
Processor memory
Processor memory
16
Hybrid Systems

• May have a top-level shared disk architecture
• Individual nodes are themselves shared memory
  systems
• Good compromise between load balancing and
  scalability

17
Web Architecture

• Need for enterprise scalability causes problems
  which can be solved by a three-tier architecture

• User interface

Client
Application Server

• Business logic
• Data processing logic

• Server-side validation
• Database access

Database Server
18
Web as a Database Platform

• Advantages
• DBMS advantages
• Simplicity
• Graphical User Interface
• Standardization
• Cross-platform support
• Transparent network access
• Scalable deployment
• Innovation

19
Web as a Database Platform

• Disadvantages
• Reliability
• Security
• Cost
• Scalability
• Limited HTML Functionality
• Statelessness
• Bandwidth
• Performance
• Immaturity of development tools

20
Approaches

• CGI
• Server Side Includes
• HTTP Cookies
• API (non-CGI gateways)
• ODBC
• Java (JDBC, JSQL, JRB)
• JavaScript, JScript
• Microsoft Active Platform (ASP, ADO, ActiveX)
• PHP (Hypertext Preprocessor)

21
Further Reading

• Connolly and Begg, Database Systems, chapters 22,
  28, 29, and appendix H.
• Ozsu and Valduriez, Principles of Distributed
  Database Systems, 2nd edition
• everything you ever wanted to know about
  distributed database systems
• Chaudri and Zicari, Succeeding with Object
  Databases, 2001.
• Elmasri, Fundamentals of Database Systems, 3rd
  edition, 1999.