1. Choosing a Computing Architecture Chapter 8

2. Architectural Requirements ScalabilityManageabilityAvailabilityExtensibility FlexibilityIntegration UserBusiness Budget Technology

3. Strategy for Architecture Definition zObtain existing architecture plans zObtain existing capacity plans zDocument existing interfaces zPrepare capacity plan zPrepare technical architecture zDocument operating system requirements zDevelop recovery plans zDevelop security and control plans zCreate architecture zCreate technical risk assessment

4. Hardware Architecture Involve all experts zNew technology zOld technology zNetworking

5. Hardware Architectures zRobust zAvailable zReliable zExtensible zScalable zSupportable zRecoverable zParallel zVLM z64-bit zConnective zOpen

6. Hardware Architectures zSMP zCluster zMPP zNUMA zHybrids use SMP and MPP

7. Evaluation Criteria Determine the platform for your needs SMPClustersNUMAMPP High Low Scalability Maturity Low High

8. Parallel Processing zParallel daily operations zShared resources - Memory - Disk - Nothing zLoosely or tightly coupled Database Application Hardware Operating system

9. Making the Right Choice zRequirements differ from operational systems zBenchmark - Available from vendors - Develop your own - Use realistic queries zScalability important

10. SMP zCommunication by shared memory zDisk controllers accessible to all CPUs zProven technology Shared memory CPU Common bus Shared disks

11. SMP zBenefits: - High concurrency - Workload balancing - Moderate scalability - Easy administration zLimitations: - Memory (cluster for improvements) - Bandwidth

12. NUMA Loosely coupled shared memory CPU Shared memory Shared memory Disk Nonuniform memory access Shared bus

13. NUMA zBenefits: - Fully scalable, incremental additions to disk, CPU, and bandwidth - Performs better than MPP - Suited for Oracle server zLimitations: - The technology is new and less proven - You need new tools for easy system management - NUMA is more expensive than SMP

14. Clusters CPU Shared memory Node 1 CPU Shared memory Node 2 CPU Shared memory Node 3 Common high-speed bus

15. Clusters zShared disk, loosely coupled zDedicated memory zHigh-speed bus zShared resources zSMP node zBenefits: - High availability - Single database concept, incremental growth zLimitations: - Scalability, internode synchronization needed z - Operating system overhead

16. MPP CPU Memory Disk

17. MPP zA shared nothing architecture zMany nodes zFast access zExclusive memory on a node zLow cost per node zScalable znCUBE configuration

18. MPP Benefits zUnlimited incremental growth zVery scalable zFast access zLow cost per node zGood for DSS

19. MPP Limitations zRigid partitioning zCache consistency zRestricted disk access zHigh memory cost per nodes zHigh management burden zCareful data placement

20. Windows NT Architecture based on the client-server model zBenefits: - Include built-in Web services - Scalability - Ease of management and control zLimitations: - Not as secure - Cannot execute programs remotely - Lack linear scalability beyond four processors - Addressing space for applications is limited to two gigabytes

21. Architectural Tiers zTiered structures: - Modular - Logical separation zDistributed structures: - Two-tier - Three-tier - Four-tier (and more)

22. Middleware Technologies for integration Gateway

23. Database Server Requirements zRobust zAvailable zReliable zExtensible zScalable zSupportable zRecoverable zParallel

24. Parallelism zDatabase zQuery zLoad zIndex zSort zBackup zRecovery

25. Further Considerations zOptimization strategy zPartitioning strategy zSummarization strategy zIndexing techniques zHardware and software scalability zAvailability zAdministration

26. Server Environments Operational servers Warehouse servers Data mart servers Open DBMS Network, relational, hierarchical Mainframe proprietary DBMS Oracle, IMS, DB2, VSAM, Rdb, Non Stop SQL, RMS Open DBMS Relational General purpose and warehouse-specific DBMS Oracle, Informix, Sybase, IBM DB2, NCR/AT&T Teradata Red Brick Open DBMS Relational and multidimensional General purpose and warehouse specific DBMS Oracle, Oracle Express, Arbor Essbase, MS SQL Server, NT

27. Parallel Processing A large task broken into smaller tasks: zConcurrent execution zOne or more processors Processor 1 Processor 2 Processor 3 Processor 4 Parallel Elapsed time Not parallel

28. Parallel Database zIncreased speed zImproved scalability zPerformance gains - Availability - Flexibility - More users Processor 1 Processor 2 Processor 3 Processor 4 Parallel

29. Parallel Query SQL code split among server processes. Sub- Query Sub- Query Sub- Query

30. Parallel Load Bypass SQL processing to speed throughput.

31. Parallel Processing zIndex Reduces the time to create zSort Allocates memory in cache efficiently zBackup Runs simultaneously from any node - Offline - Online zRecovery Runs simultaneously from redo logs zSummaries Uses the CREATE TABLES AS SELECT statement

32. Summary This lesson discussed the following topics: zOutlining the basic architecture requirements for a warehouse zHighlighting the benefits and limitations of all the different hardware architectures