1. 15 Copyright © 2006, Oracle. All rights reserved. Performance Tuning: Summary

2. Copyright © 2006, Oracle. All rights reserved. 15-2 Objectives After completing this lesson, you should be able to do the following: List best practices identified throughout the course Summarize the performance tuning methodology

3. Copyright © 2006, Oracle. All rights reserved. 15-3 Necessary Initialization Parameters with Little Performance Impact Set to the maximum number of files that can be assigned to the database DB_FILES Set to TYPICALSTATISTICS_LEVEL Set to contain at least two files on different disk drives to prevent failures from control file loss CONTROL_FILES Limit on the maximum number of cursors for each session. The setting is application- dependent; 500 is recommended. OPEN_CURSORS Location of the database in Internet dot notation DB_DOMAIN Name of the database. This should match the ORACLE_SID environment variable. DB_NAME DescriptionParameter

4. Copyright © 2006, Oracle. All rights reserved. 15-4 Important Initialization Parameters with Performance Impact Undo tablespace to be used by instance UNDO_TABLESPACE AUTO mode recommended UNDO_MANAGEMENT To be used essentially with shared server SESSIONS Maximum number of processes that can be started by that instance PROCESSES Automatic PGA management PGA_AGGREGATE_TARGET Automatically sized SGA components SGA_TARGET 8192 for OLTP and higher for DSS DB_BLOCK_SIZE To take advantage of the latest improvements of a new release COMPATIBLE DescriptionParameter

5. Copyright © 2006, Oracle. All rights reserved. 15-5 Notes only page

6. Copyright © 2006, Oracle. All rights reserved. 15-6 Sizing Memory Initially As an initial guess for memory allocation: Leave 20% of available memory to other applications. Leave 80% of memory to the Oracle instance. For OLTP: For DSS: PGA_AGGREGATE_TARGET=(total_mem*80%)*20% PGA_AGGREGATE_TARGET=(total_mem*80%)*50% SGA_TARGET=(total_mem*80%)*80% SGA_TARGET=(total_mem*80%)*50%

7. Copyright © 2006, Oracle. All rights reserved. 15-7 Database High Availability: Best Practices Use SPFILE. Multiplex redo logs. Use resumable space allocation. Create at least two control files. Enable Flashback Database. Enable block checking. Use auto-tune checkpointing. Log checkpoints to the alert log. Use database resource manager. Use Automatic Undo Management. Use Automatic Segment Space Management. Use locally managed tablespaces. Use locally managed temporary tablespaces. Enable ARCHIVELOG mode and use a flash recovery area. Set time long enough for CONTROL_FILE_RECORD_KEEP_TIME. Designate a default permanent tablespace other than SYSTEM and SYSAUX.

8. Copyright © 2006, Oracle. All rights reserved. 15-8 Undo Tablespace: Best Practices Use Automatic Undo Management. UNDO_RETENTION : –Oracle Database 10g Release 1: Set it to your flashback requirement. –Oracle Database 10g Release 2: Do not set it. Undo tablespace size: –Initial size: Small with AUTOEXTEND enabled –Steady state: Fix size using the Undo Advisor and add a 20% safe margin.

9. Copyright © 2006, Oracle. All rights reserved. 15-9 Notes only page

10. Copyright © 2006, Oracle. All rights reserved. 15-10 Temporary Tablespace: Best Practices Locally managed temporary tablespaces use a uniform extent. Extent size should be: 1 MB to 10 MB extent size: –For DSS, OLAP applications involving huge work areas –Large temporary LOBs are predominant. 64 KB or multiple less than 1 MB: –Small global temporary tables are predominant. –OLTP Temporary tablespace group increases addressability from TB to PB.

11. Copyright © 2006, Oracle. All rights reserved. 15-11 Notes only page

12. Copyright © 2006, Oracle. All rights reserved. 15-12 General Tablespace: Best Practices Use locally managed tablespaces with auto- allocate extents policy. Use Automatic Segment Space Management (ASSM). Use online segment shrink to eliminate internal fragmentation. Periodically review the results of the Automatic Segment Advisor. Monitor tablespace space usage using server- generated alerts. Size of extents matter more than the number of extents in the segments.

13. Copyright © 2006, Oracle. All rights reserved. 15-13 Internal Fragmentation Considerations Watch for: –Bad choices of PCTFREE and PCTUSED for heap segments –Bad choices of PCTVERSION and RETENTION for LOB segments –Low density of data in segment –Direct loads followed by deletes (no inserts) –Indexes on tables with random updates and deletes with no further inserts Remedy: –Online segment shrink –Online redefinition – MOVE operations

14. Copyright © 2006, Oracle. All rights reserved. 15-14 Notes only page

15. Copyright © 2006, Oracle. All rights reserved. 15-15 Block Size: Advantages and Disadvantages Index leaf block contention (OLTP) Good for sequential access Waste cache space with random access of small rows Lower overhead: more room for data Larger For large rows (chaining)Reduce block contention (OLTP) Relatively high overhead (block header) Small rows with lots of random access Smaller DisadvantagesAdvantagesBlock Size

16. Copyright © 2006, Oracle. All rights reserved. 15-16 Sizing Redo Log Files Size of redo log files can influence performance. Larger redo log files provide better performance. Generally, redo log files should range between 100 MB and a few gigabytes. Switch redo log file at most once every twenty minutes. Use the Redo Logfile Size Advisor to correctly size your redo logs.

17. Copyright © 2006, Oracle. All rights reserved. 15-17 Automatic Statistics Gathering STATISTICS_LEVEL = TYPICAL | ALL Statistics are gathered by the predefined GATHER_STATS_JOB job. This job implicitly determines the following: –Database objects with missing or stale statistics –Appropriate sampling percentage necessary to gather good statistics on those objects –Appropriate columns that require histograms and the size of those histograms –Degree of parallelism for statistics gathering –Prioritization of objects on which to collect statistics

18. Copyright © 2006, Oracle. All rights reserved. 15-18 Automatic Statistics Collection: Considerations You should still manually gather statistics in the following cases: –After bulk operations –When using external tables –To collect system statistics –To collect statistics on fixed objects Prevent automatic gathering for volatile tables: –Lock with statistics for representative values –Lock without statistics implies dynamic sampling.

19. Copyright © 2006, Oracle. All rights reserved. 15-19 Commonly Observed Wait Events commits + rollbacks from V$SYSSTAT, Disks. Slow I/O, un-batched commits Over-commit, I/O Log file sync V$SYSSTAT redo buffer allocation retries, disk Small buffer, slow I/OLog buffer I/Olog buffer space V$SQLAREA parse calls, child cursors SQL parsing/sharingLatchesLibrary cache waits V$ENQUEUE_STAT Depends on enq typeLocksEnqueue waits (enq:) V$SQLAREA disk reads. V$FILESTAT read time Poorly tuned SQL, Slow I/O system I/O, SQL Tuning db file scattered read, db file sequential read Write time using OS stats. Buffer cache stats Slow DBWRBuffer cache, DBWR, I/O free buffer waits V$SESSION (block) while issue is occurring Depends on buffer type. PK index and seq. Buffer cache, DBWR buffer busy waits ExaminePossible causeAreaWait Event

20. Copyright © 2006, Oracle. All rights reserved. 15-20 Additional Statistics Use automatic undo, tune the workload How many rollbacks done for consistent read purposes Consistent changes Checkpoints, DBWR, or archiver activity should be tuned, larger log files How many times a server process had to wait for space in the online redo log Redo Log Space Requests Use automatic undo, tune the workload How many blocks are read in Consistent Read mode Consistent gets ReorganizeMigrated or chained rows Table Fetch by Continued Row Recommended actionDescriptionStatistic name

21. Copyright © 2006, Oracle. All rights reserved. 15-21 Top 10 Mistakes Found in Oracle Systems 1.Bad connection management 2.Bad use of cursors and shared pool 3.Bad SQL 4.Use of nonstandard initialization parameters 5.Getting database I/O wrong 6.Redo log setup problems 7.Serialization of data blocks in the buffer cache 8.Long full-table scans 9.High amount of recursive SQL 10.Deployment and migration errors

22. Copyright © 2006, Oracle. All rights reserved. 15-22 Notes only page

23. Copyright © 2006, Oracle. All rights reserved. 15-23 Summary In this lesson, you should have learned how to: Create your initial database following the best practices identified throughout the course Summarize the performance tuning methodology

24. Copyright © 2006, Oracle. All rights reserved. 15-24