Yep, we had to do this too, before upgrading to a GE release (UGE, in our
case) that supports cgroups. cgroups allow you to limit on actual physical
pages used by a job, rather than address space request like h_vmem.

On Thu, Jun 02, 2016 at 05:25:30PM +0200, Julien Nicoulaud wrote:
> I have had issues in the past with Java jobs and h_vmem. When using a large
> initial heap size, SGE would kill the java process at startup when it is
> allocating its memory. You may want to lookup the MALLOC_ARENA_MAX
> environment variable.
> 
> We ended up using s_vmem instead, and handling it in our job scripts.
> 
> 2016-05-09 18:31 GMT+02:00 Alex Chekholko <ch...@stanford.edu>:
> 
> > IIRC this bug had to do with setting your "consumable" value to "YES" vs
> > "JOB".  I think it only affected you if you have your consumable value per
> > job instead of per slot.
> >
> > I have now:
> >
> > #name               shortcut   type        relop requestable consumable
> > default  urgency
> >
> > h_vmem              h_vmem     MEMORY      <=    YES         YES 1G       0
> >
> > We used to use "JOB" in that consumable column.
> >
> >
> > On 05/09/2016 02:13 AM, nico.faer...@id.unibe.ch wrote:
> >
> >> Hi,
> >>
> >> We are running OGS/GE 2011.11p1. Concerning the problem that for certain
> >> multi-core jobs the reported h_vmem value is much higher than the true
> >> amount of memory used by the job, can anyone propose a solution to this
> >> problem, e.g. working patch, ??? How do you deal with this problem? We
> >> want to keep h_vmem consumable to prevent jobs from overbooking a node.
> >>
> >> Thank you,
> >> Nico
> >>
> >>
> >> _______________________________________________
> >> users mailing list
> >> users@gridengine.org
> >> https://gridengine.org/mailman/listinfo/users
> >>
> >>
> > --
> > Alex Chekholko ch...@stanford.edu
> >
> > _______________________________________________
> > users mailing list
> > users@gridengine.org
> > https://gridengine.org/mailman/listinfo/users
> >

> _______________________________________________
> users mailing list
> users@gridengine.org
> https://gridengine.org/mailman/listinfo/users


-- 
-- Skylar Thompson (skyl...@u.washington.edu)
-- Genome Sciences Department, System Administrator
-- Foege Building S046, (206)-685-7354
-- University of Washington School of Medicine
_______________________________________________
users mailing list
users@gridengine.org
https://gridengine.org/mailman/listinfo/users

Reply via email to