6. Problem Behind Manual Tuning
1. Abundant expertise.
2. Not time efficient
3. Misdirected
4. Methodology without having
holistic view
5. Regenerate the problem
6
8. Database Time
• Common Currency
Figure 1:Database time from a single user’s perspective
8
9. Automatic Database Diagnosis Monitor
(ADDM)
• Oracle 10g self-managing framework
• Goal: Find root-causes of problems
• Which component of database consume significant time?
• What is the top issue affecting overall throughput?
9
10. Data for ADDM
• Collect Data from each node of DB Time-graph
• Active Session History
10
14. Data: Non-numeric in nature
Full log of database setting (Not changes frequently)
• Size of memory components
• Number of CPUs using by the system
• Special query optimizer setting
14
15. - AWR
- ADDM
- Recommendation
- Oracle 10g Advisors
- In memory
15
Self-Managing Database Framework
Figure 2: ADDM and the Self-Managing Database framework
17. Observe Phase
• Automatic and continues
• Collect relevant measurements and statistics
• Precise timing information
• Automatic Workload Repository (AWR): the main component
17
18. AWR
• Store of performance data
• The database captures system & stores performance data in AWR
• Each collection as Snapshot
• ADDM takes two most recent snapshots
18
19. Diagnose Phase
• Fetch data from AWR
• Analyze Data
• Set of advisors
• ADDM itself an Advisor
• SQL tuning Advisor
• Segment Advisor
• Memory Advisor
- If any advisor consumes substantial amount of time to analysis, ADDM
takes part.
19
20. Resolve Phase
• Apply recommendation automatically or manually
• Considered for future performance measurement
• Closes iteration of particular tuning loop
• Tuning loop till desired performance level
20
21. Experiments and Results
• Hard to quantify
• Need survey
• Real ADDM usage: Experience of Qualcomm
• Updated Oracle version 8.1.7.4 to 10g.
• Significant Performance Fall
• ADDM recommendation: SQL update statement causes 90% system load
• SQL Tuning Advisor: Create index
Reason: A patch was missing in updated version
21
22. Conclusion
• Comprehensive tuning solution
• Database time of various component
• Marginal effect of component on Database
• Omit reconstruct the problem
• DBTime-graph isolate the root cause
• ADDM provides Recommendation
22