Page History
...
- single-core runs over 380*80 events
- 80-core job is running on 380 events per core
- use ds.break_after(380*80) in stead of termination of each. rank after. reaching 380 events.
Better start-stop time definition:
- skip 1-st event, because of time consumption overhead due to loading of calib constants
- start time is preserved for the 1-st rank gettein 2-nd event
- stop time for the last finishing rank
Summary
Total processing time, sec, of 380*80 events
Ratio is fare from 80...
| number of cores | old algorithm | ←Ratio t1/t80 | new algorithm v2 | ←Ratio t1/t80 | new algorithm v2 with loop | ←Ratio t1/t80 |
|---|---|---|---|---|---|---|
1 better start-stop | 11102 | 39.5 27.1 | 8425 | 48.0 51.5 | 8570 | 50.2 57.8 |
| 80 | 281 | 1 | 175.4 | 1 | 170.6 | 1 |
80 - better start-stop time def, times for 3 jobs: average: | 418.3 408.4 401.0 409.2 | 1 | 157.4 165.1 168.4 163.6 | 1 | 148.6 148.2 147.7 148.2 | 1 |
References
- Scaling behavior of psana1 - Part 1 - det.calib method in multicore processing with mpi
- Scaling behavior of psana1 - Part 2 - test with command perf stat
...
Overview
Content Tools