[slurm-users] Job runtime

[Mahmood Naderan]

Hi Chris,
I have been confused with the cpu runtime values in the sacct. For a
multinode mpi job, I see these values

[mahmood at rocks7 ~]$ sacct --format=jobid,user,cputime,elapsed,totalcpu,ncpus
       JobID      User    CPUTime    Elapsed   TotalCPU      NCPUS
------------ --------- ---------- ---------- ---------- ----------
24             mahmood   23:01:25   00:11:25   22:57:37        121
24.batch                 06:05:20   00:11:25   06:04:16         32
24.0                     00:34:15   00:11:25   16:53:21          3


The questions are:
1- Why cputime is greater than totalcpu?
2- Considering cputime, the sum of 24.batch and 24.0 is not equal to 24. Why?

Thanks for your help.


Regards,
Mahmood




On Sun, Mar 18, 2018 at 12:08 PM, Chris Samuel <chris at csamuel.org> wrote:
> On Sunday, 18 March 2018 4:30:34 PM AEDT Mahmood Naderan wrote:
>
>> Thanks for the exaplnation Chris. I will read cgroup.
>
> My pleasure!   The Slurm docs on it are here:
>
> https://slurm.schedmd.com/cgroups.html
>
> I've been using cgroups for all three abilities (process tracking, task
> management and accounting) since 2013 and found it works really well.
>
> All the best,
> Chris
> --
>  Chris Samuel  :  http://www.csamuel.org/  :  Melbourne, VIC
>
>