B1_Soil Erosion Performance Assessment
This page should describe Performance Assessment Tests performed for this stand alone feature and should provide links to all the result pages.
Summary
Testing was performed on Cori-KNL to assess performance of the coupled ELM and soil erosion code, using 192 cores and recording total time for completion, and simulated four years (2001–2004).
Performance Test 1
Performance Test 1: Performance on Cori-KNL and Compy
Date last modified: Jun 14, 2019
Contributors: @Zeli Tan
Provenance: 360x720 cru grid, tanzeli/lnd/elm-erosion-master branch
Results:
Hash tag | Machine | Processor Count | minutes per simulated year | simulated years per wall-clock day | minutes to initialize |
|---|---|---|---|---|---|
a4c3ad0 (development) | Cori-KNL | 192 | 227 m 45 s | 6.3 simulation years | 4 m |
0b9a938 (development) | Compy | 160 | 111 m 30 s | 12.9 simulation years | 0.8 m |
59ce0c8 (baseline) | Compy | 160 | 109 m 43 s | 13.1 simulation years | 0.8 m |
Comments
@Robert Jacob, thanks for your comment. Could you be more specific? Should I run two simulations (one with my changes and another without my changes) on Compy? Or more are preferred?
Yes 2 simulations, one with and one without, on Compy should be fine. You only need to do more if you see results you don't expect.
For this Assessment to be helpful, include a measurement at the start of the branch, before any changes were made. You may need to do several runs because cori has a lot of variance. Using anvil or compy might be better.