[torqueusers] Torque not counting cputime for all nodes

Garrick Staples garrick at clusterresources.com
Fri Aug 11 19:55:17 MDT 2006


On Fri, Aug 11, 2006 at 02:46:02PM +0200, Martin Schafföner alleged:
> Hi all,
> 
> I noticed after upgrading to 2.1.1 and lately to 2.1.2 that Torque does not 
> count the used cputime correctly. I.e., I have a job running on both CPUs on 
> a number of SMP nodes at full cpu throttle, so cputime should be 
> approximately ncpus*walltime, but it's always only 2*walltime. And yes, 
> processes are launched through the TM interface.

How long are these running for?  Maybe what you are seeing is the lag in
polling time on sisters?

Maybe decrease check_poll_time in your pbs_mom config?



More information about the torqueusers mailing list