[torqueusers] how does (pv)mem limits work

[Ake Sandgren]

On Thu, 2010-09-16 at 09:15 -0400, Glen Beane wrote:
> On Thu, Sep 16, 2010 at 6:37 AM, Ramon Bastiaans
> <ramon.bastiaans at sara.nl> wrote:
> >  I wonder how pvmem is supposed to work, because when I'm testing it does
> > not behave as I would expect it to.
> >
> >  * if I set resources_max.pvmem=8192m, the environment gets no ulimit. Ok
> > fine, resources_default then;
> >  * if I set resources_default.pvmem=8192m and submit -lnodes=1:ppn=1 the
> > environment gets an ulimit (-v) of 8192 vmem (good)
> >  * if I set resources_default.pvmem=8192m and submit -lnodes=1:ppn=2 the
> > environment gets an ulimit (-v) of 8192 vmem (weird?)
> 
> I believe each process would have a ulimit of 8192, giving your job
> 16GB total.  If you want to submit a threaded job that uses all 16GB
> within a single process then I think you would have to request 16GB,
> but I think the scheduler might then look for a node with 2 * 16GB
> (singe ppn=2, and it doesn't know you are going to be running a single
> process with multiple threads).  Maybe someone with some more
> experience using pvmem defaults can answer...

You are quite right. That's exactly how it works.

-- 
Ake Sandgren, HPC2N, Umea University, S-90187 Umea, Sweden
Internet: ake at hpc2n.umu.se   Phone: +46 90 7866134 Fax: +46 90 7866126
Mobile: +46 70 7716134 WWW: http://www.hpc2n.umu.se