Collecting all successfully completed xdsapp jobs in ponline_p11 partition where the runtime was longer than 60 seconds and shorter than 30 minutes, shows that the vast majority of jobs have a runtime of 6-7 minutes on P11 nodes max-p3a018,9.

P11 can run 4 concurrent jobs, So two nodes hence offer the capability to process 40 typical datasets per hour (assuming a typical runtime of 6 minutes).

Processing in ponline_p11_com has too few jobs for good statistics. A simple benchmark go

  • max-itt001 =~ 0.89*max-p3a018 / 2
  • max-itt001 =~ 0.59*max-p3a018 / 2

max-itt(g)001 can only take one xdsapp job at a time.

The two nodes in the ponline_p11_com partition hence offer the capability to process ~15 typical datasets per hour, in comparison to 40 datasets/h in the ponline_p11 partition.


Benchmarks for AMD EPYC

CPUJobIDNodeTotal CoresRAMAvg runtimeRatio  / EPYC 7642Jobs/nodeDatasets/hour/node
EPYC 7642 9359589max-p3a018192256G343.7s1.0220
EPYC 73029359590max-itt00164512G385.7s0.8919
Silver 42109361012max-ittg00140384G576.7s0.5916
EPYC 73519362984max-mllab-amd013260G672.7s0.5115
EPYC 7H129363618max-cdcs0012561024G238.7s1.443~40
EPYC 7F529363636max-cdcs002641024G382.7s0.9019
EPYC 75429363738max-exfl273128512G444.0s0.771-2~10
EPYC 74029364011max-wn06496512G455.7s0.7518
EPYC 75529364437max-mpag0011921024G408.0s0.84216
EPYC 72629364439max-mpag00516256G769.3s0.4414

Job Profile