[torqueusers] reading the "Time Use" in qstat
garrick at clusterresources.com
Thu Sep 7 12:57:29 MDT 2006
On Thu, Sep 07, 2006 at 11:38:23AM +0100, David Golden alleged:
> On Thursday 07 September 2006 11:16, bill wrote:
> > Hello
> > I' m not sure what means the "Time Use" column in qstat.
> > Job id Name User Time Use S Queue
> > ------------------- ---------------- ---------------- -------- - -----
> > 128.m1 possession ghon 44:35:09 R batch
> > 157.m1 ...ession_parall ghon 00:00:00 R batch
> > The jobs are pretty much equals. Both jobs are running for two days now.
> > Why there is one consuming Time Use, and another one not?
> At some point I think torque qstat started reporting CPU time used
> in that column, rather than job wall time used. Maybe it always did, and
> I'm (ahem) "misremembering" somehow. So maybe the latter job isn't
> using much cpu time. (Some old versions of torque also had time reporting
> bugs, but chances are you're not running one of 'em, especially since
> one of the jobs has a real-looking figure...)
'qstat' has always had CPU time. 'qstat -a' has always had walltime.
> It's of course useful to know when users are arguably abusing cluster nodes
> they've reserved by mis-sizing their jobs leading to CPU underuse, but in the
> end we "bill" (currently only notionally in our academic case, but still) 'em
> for used wall time, not for actual used cpu time (they're typically blocking
> other people from using the node whether or not they're using the cpu
> properly...), so I'd rather the qstat default just showed elapsed wall time
> of job so far...
> Might be time for me to read the qstat man page again... :-)
The manpage does detail each column in the various display modes. The
standard mode is specified in POSIX and SuS, so we can't really change it.
More information about the torqueusers