[torqueusers] (sub)queue configuration questions

[Garrick Staples]

On Thu, Jan 29, 2009 at 04:54:38PM -0800, Jim Kusznir alleged:
> Hi all:
> 
> I have a cluster that I (currently) want two queues on.  1 queue is
> the default, and it will use all nodes in the cluster for jobs as
> needed.  It has a 24 hr walltime limit.
> 
> The second queue is the "long" queue.  It has no walltime limit, but I
> want to limit how many cores of the cluster it will use to 64 (1/3 of
> my cluster).  Ideally, users may queue many jobs totalling well over
> it, but even of the resources are available, I do not want jobs from
> this queue to ever use more than 64 cores.  If I need to lock it down
> to a speciifc set of nodes, that's fine.  However, if someone queues a
> large job in the default queue that needs those cores, I want "normal"
> scheduling to take place (i.e, torque will schedule the job to execute
> once those cores are available, etc).  The idea is to ensure that at
> least 2/3 of the cluster is available for jobs at least every 24
> hours, but for those users who can't break up their job, they can
> still run, but can't take over a large portion of the cluster.
> 
> I've tried my best guesses based on the torque manual, but so far my
> attempts have had very undesirable effects:
> 
> 1) every job (including ones explicitly wanting 1 node) get changed to 64;
> 2) jobs requiring more than 64 are still run

You can't do this in torque by itself.  But the maui config is pretty easy:
CLASSCFG[long]          QDEF=long QLIST=long
QOSCFG[long]            MAXPROC=64

-- 
Garrick Staples, GNU/Linux HPCC SysAdmin
University of Southern California

See the Prop 8 Dishonor Roll at http://www.californiansagainsthate.com/

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20090129/b784f916/attachment.bin