Availability and Maintainability Benchmarks A Case Study of Software RAID Systems
Overview
Part I
Outline: Availability Benchmarks
Why benchmark availability?
Step 1: Availability metrics
Step 2: Measurement techniques
Step 3: Reporting results
Case study
Benchmark environment
Benchmark environment: faults
Single-fault experiments
Multiple-fault experiments
Comparison of systems
Transient error handling
Transient error handling (2)
Reconstruction policy
Reconstruction policy: graphical view
Reconstruction policy (2)
Double-fault handling
Availability Conclusions: Case study
Conclusions: Availability benchmarks
Availability: Future opportunities
Part II
Outline: Maintainability Benchmarks
Motivation
Metrics & Approach
Methodology
1) Build a task taxonomy
2) Measure a task’s cost
3) Measure task frequencies
4) Apply a cost function
Case Study
Experimental platform
Experimental procedure
Experimental procedure (2)
Experimental procedure (3)
Sample results: time
Analysis of time results
Sample results: learning curve
Discussion
Discussion (2)
Early reactions
Future Directions: Maintainability
Searching for feedback...
Backup Slides
Approaching availability benchmarks
Example Quality of Service metrics
System configuration
Single-fault results
Behavior A: no effect
Behavior B: lost redundancy
Behavior C: automatic reconstruction
Behavior D: system failure
System comparison: single-fault
Example multiple-fault result
Multi-fault results
Multi-fault results (2)
Email: abrown@cs.berkeley.edu
Home Page: http://istore.cs.berkeley.edu/
Download presentation source