Related to this, it would a good idea if a job's memory requirements
were passed to pbs...at present its not possible for the scheduler
to do its job properly.
On Wednesday 27 April 2005 05:42, Gordon, JC (John) wrote:
> Fraser, probably before your time but Glasgow was one of the first to
> point out this problem with Maui. You need to set the default high. This
> is a drag but no obvious way round. It is a long-standing requirement on
> WMS people, not sure if gLite WMS addresses it.
>
> This has reared its head again recently as long defaults prevent PBS (or
> whatever) backfilling spare cpus as it tries to gather a number of free
> cpus for an MPI job.
>
> John
>
> > -----Original Message-----
> > From: Testbed Support for GridPP member institutes
> > [mailto:[log in to unmask]] On Behalf Of Fraser Speirs
> > Sent: 26 April 2005 15:46
> > To: [log in to unmask]
> > Subject: Walltime for LCG jobs
> >
> > All,
> >
> > Please excuse the vagueness of this question, it concerns the part of
> > our farm that I don't deal with directly (the PBS system).
> >
> > I'm told that LCG jobs appear in PBS with an estimated wall time of
> > one minute and that this means PBS will always kill the job after a
> > minute, regardless of the job's actual requirements.
> >
> > Is this something that other people have had to work around and, if
> > so, what's considered the best approach?
> >
> > Thanks,
> > Fraser
> > --
> > Technical Coordinator - ScotGrid
> > a: Room 481, Kelvin Building, University of Glasgow
> > w: http://ppewww.ph.gla.ac.uk/~fspeirs/ t: 0141 330 6439
--
------------------------------------------------------------------------------
| |
| Dr. Alex Martin |
| e-Mail: [log in to unmask] Queen Mary, University of London, |
| Phone : +44-(0)20-7882-5033 Mile End Road, |
| Fax : +44-(0)20-8981-9465 London, UK E1 4NS |
| |
------------------------------------------------------------------------------
|