[Moabusers] large number of jobs queued

[Glen Beane]

On 9/26/06, Justin Bronder <jsbronder at gmail.com> wrote:
> On 9/25/06, Thomas G. Raisor <thunder at et.byu.edu> wrote:
> > Hi,
> >
> > I have about 25,000 jobs in my torque queue right now, but moab is only
> > seeing roughly the first 4100 (using showq).  Jobs not shown with showq
> > give the following error when I do a checkjob on them.
> >
> > ERROR:  cannot locate job 'jobid'
> >
> > I can run the jobs with qrun with no problems. This is a vanilla install
> > of moab - are there defaults parameters I need to increase? I was using
> > an older patch release of moab 4.5, updated to the latest and get the
> > same behavior. Could it be torque not communicating its jobs to moab?
> > torque version is 2.1.0 - (yes, I know I am a little out of date, but I
> > haven't had any problems until now.)
> >
> > Suggestions?
> >
> > Tom
> >
> > _______________________________________________
> > moabusers mailing list
> > moabusers at supercluster.org
> > http://www.supercluster.org/mailman/listinfo/moabusers
> >
>
>
>
> It should be seeing the first 4096 jobs actually.  Anyways, you can adjust
> the
>  default limits, which is something I had to do for partitions.  Refer to
> the
>  following page and the MAXJOB variable.
>
> http://www.clusterresources.com/products/mwm/docs/a.ddevelopment.shtml
>
>  -Justin.

Also note that although by default Moab would only "see" the first
4096 jobs that just means it will only consider the first 4096 jobs
while scheduling.  As these jobs finish,  it will consider the jobs
that it was not previously "seeing".   So it would be like a 'sliding
window' of size 4096, and moab will not consider anything outside of
this window.  I'll need moab devs to confirm this, but I would think
that making this window too large could result in a performance
penalty without guaranteeing a significant improvement in the overall
scheduling efficiency.