CompSci 296.2 Self-Managing Systems Shivnath Babu

Download Report

Transcript CompSci 296.2 Self-Managing Systems Shivnath Babu

CompSci 296.2
Self-Managing Systems
Shivnath Babu
Today
• IBM’s LEarning Optimizer (LEO)
• Oracle’s Automatic Database Diagnostic Monitor
2
Background
• Declarative queries
• Query plans
• Query optimizer
– Indexes, index-AND’ing, materialization
– Statistics, e.g., Histograms, quantiles (percentile)
• Assumptions made by query optimizer
– Uniformity, inclusion, independence
3
LEO
• Feedback-based learning
– Inter-query
4
LEO
• Feedback-based learning
– Intra-query
5
Challenges
• Stability
• Detecting a problem
• Deciding to respond to a problem
• Dealing with bias, incompleteness, uncertainty
6
Oracle ADDM
• Root-cause identification
• Recommend fix
• Characterize improvement
7
Oracle ADDM
• DB-Time
– Need common unit; DB-Time Vs. Wall-Clock time
• DB-Time graph
• Data collection
– Time
– Active Session History
• Observe, Diagnose, Resolve
• Experimental evaluation
8
Discussion
• Relationship between the two papers?
• Differences between the papers?
• Can ADDM answer the What-if question on Page 3?
– Dealing with concurrency
• Accounting for DB-Time
• Extending DB-Time
• Experimental evaluation
• Can we extend ADDM approach to 3-tier system?
• Stability
9