[HTCondor-users] Discrepancy between RemoteCpu and CumulativeRemoteCpu in Parallel Universe jobs

[Carles Acosta <cacosta@xxxxxx>]

Dear all,

We are currently running a project that executes parallel jobs using MPI via HTCondor's Parallel Universe. The jobs are completing successfully, so execution is not the issue. However, we have noticed a significant discrepancy between CumulativeRemoteCpu and RemoteCpu, and we do not understand why, given that the jobs have only started once (NumJobStarts = 1). The same issue applies to RemoteSysCpu.

Here you have an example:

# condor_history -limit 1 1505513.0 -af JobUniverse NumJobStarts RemoteUserCpu CumulativeRemoteUserCpu RemoteSysCpu CumulativeRemoteSysCpu RemoteWallclockTime MaxHosts RequestCpus
11 1 7596234.0 89264.0 79249.0 6601.0 26827.0 2 288

The CumulativeRemoteUserCPU is significantly lower than the RemoteUserCpu. Furthermore, the RemoteUserCpu seems to only be accounting the cpu consumed by one of the two nodes (7596234.0/288 ~Â26376). Is this a known issue when using JobUniverse 11?

Thank you in advance for your help!

Cheers,

Carles

--

Carles Acosta i Silva

PIC (Port d'Informacià CientÃfica)

Campus UAB, Edifici D

E-08193 Bellaterra, Barcelona

Tel: +34 93 581 33 08

Fax: +34 93 581 41 10

http://www.pic.esÂ

AvÃs - Aviso - Legal Notice: Âhttp://legal.ifae.es