DGX Nightly Benchmark run 20210414

[quasiben]

Benchmark history

<img width="641" alt="Benchmark Image" src="https://raw.githubusercontent.com/quasiben/dask-scheduler-performance/benchmark-images/assets/dgx-20210414-benchmark-history.png">

Raw Data

<Client: 'tcp://127.0.0.1:33011' processes=10 threads=10, memory=503.79 GiB> Distributed Version: 2021.02.0+36.g4d792ef3 simple 5.364e-01 +/- 2.341e-02 shuffle 2.267e+01 +/- 1.272e+00 rand_access 1.039e-02 +/- 1.135e-03 anom_mean 9.903e+01 +/- 1.104e+00

Raw Values

simple [0.53973051 0.51369161 0.58278502 0.51988353 0.5223909 0.51060436 0.56974273 0.55412855 0.52701474 0.52403028] shuffle [21.40838883 21.39001123 21.99433303 21.82835566 21.85750325 22.26745541 25.32918345 22.72368488 24.519427 23.36290505] rand_access [0.00747297 0.011455 0.01069607 0.0098767 0.01198526 0.01091527 0.01028906 0.0104905 0.01024728 0.01048931] anom_mean [100.41342478 98.3349343 97.96303732 98.69094799 98.87538351 100.45040971 97.31347765 99.63593017 98.06277313 100.53165452]

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/20210414-sched-graph.png">

[quasiben]

This benchmark was executed with @madsbk's PR: https://github.com/dask/distributed/pull/4699