On Tue, Mar 08, 2005 at 07:41:44PM +0100 or thereabouts, Maarten Litmaath wrote:
> Charles Loomis wrote:
>
> >Hello,
> >
> >I've managed to get a reasonable configuration of torque, maui, mpich,
> >and mpiexec working at LAL. The configuration fixes the problem you
> >describe below and also adds support for mpiexec. It uses newer
> >versions of torque and mpich than are standard in the LCG/EGEE
> >distribution. The newer versions are running at LAL and work well.
> >
> >There is a short readme on the configuration which needs to be changed
> >here:
> >
> >http://grid05.lal.in2p3.fr/mpi/readme.txt
> >
> >and the rpms can be found here:
> >
> >http://grid05.lal.in2p3.fr/mpi/
>
> Should we use those for LCG-2_4_0? Steve, what about your version?
I've been wanting to look at the newer version of torque here as
well but have not got around to it. In particular I wanted
to add --disable-rpp which configures the mom communication to use
tcp rather than "reliable packet protocol" whatever that is. We see
timeouts though that is probably just a function of farm size.
I can't really test a new version unless we have a powercut which
is what I normally use for upgrading torque.
There is a request that I left a library out from the -devel
package as well which is needed for MPICH.
It looks like I should build new packages with Cals notes and I would
like to '--disable-rpp'. The experiences of this version are generally
good on the torque mailing list.
This would require that there be no running jobs for the upgrade. Is
2.4.0 a significant enough upgrade for this to be required anyway?
>
> >MPICH and mpiexec have been compiled to use p4 by default which is
> >probably what you'll want.
As for the MPICH and mpiexec then this is something I don't know about.
> >
> >Cheers.
> >
> >Cal
> >
> >
> >
> >Klaere Cassirer wrote:
> >
> >>Hello,
> >>I have questions about MPICH jobs. Our site runs LCG2.3 on SL3 with the
> >>maui
> >>scheduler and pbs.
> >>- Does anybody know how to specify a NODENUMBER greater than the
> >>number of
> >>WNs on a site, which has more than 1 CPU per WN? I want to use all
> >>available
> >>CPUs (here: 2 CPUs per WN).
> >>- How can I tell maui to reserve x nodes for me, but only one CPU per
> >>node?
> >>
> >>In LCG2.2 times I specified nodenumber=x, then ran a script which
> >>started 2
> >>mpich-processes on each of the x nodes. But with maui, I get x CPUs
> >>(every
> >>2nd on the same WN), but not more.
> >>
> >>Regards
> >>
> >>--
> >>Klaere Cassirer
> >>Fraunhofer-Institute for Algorithms and Scientific Computing (SCAI)
> >>Schloss Birlinghoven
> >>D-53754 Sankt Augustin
> >>
> >>Tel: +49 - 2241 - 14 - 2758
> >>Fax: +49 - 2241 - 14 - 42758
> >>E-mail: [log in to unmask]
> >>Internet: http://www.scai.fraunhofer.de
> >>
>
>
--
Steve Traylen
[log in to unmask]
http://www.gridpp.ac.uk/
|