DGX Nightly Benchmark run 20210506

[quasiben]

Benchmark history

<img width="641" alt="Benchmark Image" src="https://raw.githubusercontent.com/quasiben/dask-scheduler-performance/benchmark-images/assets/dgx-20210506-benchmark-history.png">

Raw Data

<Client: 'tcp://127.0.0.1:33535' processes=10 threads=10, memory=503.79 GiB> Distributed Version: 2021.04.1+13.g7bfd1ae2 simple 6.064e-01 +/- 3.416e-02 shuffle 2.292e+01 +/- 1.163e+00 rand_access 6.262e-03 +/- 3.449e-03 anom_mean 1.02e+02 +/- 1.511e+00

Raw Values

simple [0.5961892 0.58853566 0.63636864 0.56096533 0.6282183 0.60444839 0.59633865 0.57874093 0.68736619 0.58690996] shuffle [22.06757594 21.62102829 21.79619415 22.00976714 23.02467824 22.95085546 23.0562991 22.96852037 24.01314765 25.71277775] rand_access [0.01035976 0.00449991 0.00382113 0.00360107 0.00788093 0.00334573 0.00546885 0.0040071 0.00512981 0.01450133] anom_mean [103.66085386 100.33133125 102.70150876 99.90952873 101.19357276 101.73470926 101.41326833 104.74782515 100.69413257 103.44724941]

Dask Profiles

• Shuffle Profile
• Random Access Profile
• Simple Profile
• Anom Mean

Scheduler Execution Graph

<img width="641" alt="Sched Graph Image" src="https://raw.githubusercontent.com/quasiben/dask-scheduler-performance/benchmark-images/assets/20210506-sched-graph.png">

[jakirkham]

This was run @gjoseph92's changes in PR ( https://github.com/quasiben/dask-scheduler-performance/pull/130 ). Hopefully this gives us a common baseline going forward