Example: Patterns and Performance
In the traced execution, Pious ran across five file-systems (one daemon per file-system plus one master) with two test processes running a set of tests: 8 processes, 33845 events
Identified three high-level patterns
- SCATTER: occurred 830 times 15.65 secs
- GATHER: occurred 537 times 12.54 secs
- Distributed REQUEST-REPLY: occurred 201 time 11.13 secs
- (all times taken on IBM RS6K, model 25T, system+user time)
Longer search strings speed up performance
Elapsed times were all around 45 seconds, making approach feasible for interactive use