[torqueusers] Xeon submit host for both Xeon and Opteron nodes?

Justin Bronder jsbronder at gmail.com
Fri Nov 24 08:09:39 MST 2006


A quick fix is to edit all of the mom_priv/config files and add "arch xeon"
at the end of the file then restart the moms.  When you submit you'll
have to add "-l arch=xeon".

Of course, then your users have to do this each time and there's no
reason for the separate queues.  Perhaps a better way is to use nodesets.
http://www.clusterresources.com/products/maui/docs/8.3nodesetoverview.shtml

For instance in server_priv/nodes
xeonnode np=2 pXEON

Then in maui:
FEATUREPARTITIONHEADER p
NODESETATTRIBUTE FEATURE
NODESETLIST pXEON pAMD

CLASSCFG[ruche] DEFAULT.NODESET=ONEOF:FEATURE:XEON

This separates your cluster into partitions, so jobs cannot cross into xeon
and
amd machines, and also does the limiting of ruche jobs to only xeon nodes.

Even better, you don't have to edit all the mom config files, just the
server and
maui configs.

-Justin.

On 11/23/06, Jacques Foury <Jacques.Foury at math.u-bordeaux1.fr> wrote:
>
> Ryan Thompson a écrit :
> > Hi Mike,
> >
> > We currently use an Intel Xeon submit node and all our compute nodes
> > are amd opterons.
> >
> > If you use separate queues it should be fine.
> >
> > Cheers
> > --
> > Ryan Thompson,
> > HPC Systems Administrator
> > ryan at zymeworks.com
>
> Hi Ryan and all others.
>
> Here we have a problem we've not seen yet.
>
> I've parted the cluster with 11 Xeon nodes (named ruche) and 14 Opteron
> nodes (named callas).
>
> I've made pure opteron queues and pure xeon queues with :
>
> - in the nodes file
> xeonname np=2 xeon ruche
> ...
> opteronname np=2 opteron callas
>
> - in the queues definitions
> set queue ruche resources_default.neednodes = ruche
> set queue callas resources_default.neednodes = callas
>
> - in the MAUI config
>
> CLASSCFG[callas] MAXJOB=48
> CLASSCFG[ruche] MAXJOB=22
>
> And we thought all was all right !
>
> BUT if we submit more than 48 jobs in the callas queue, the 49th job is
> submitted to a xeon node !!!
>
> How can I ensure this never happens ?
>
> (what I found is :
> # checknode xeonname
>
>
> checking node xeonname
>
> ...
>
> Features:   [xeon]
> Attributes: [Batch]
> Classes:    [callas 2:2][ruche 2:2][batch 2:2]
>
>
> how comes the "Features" tell only "xeon" "ruche" but we can see
> "callas" in the classes ?????)
>
>
> Thanks for any help...
>
>
> --
>
> Jacques Foury
> Ingénieur d'études
> Institut de Mathématiques de Bordeaux
> Université Bordeaux 1 / CNRS
> Tel : 05 4000 69 56
> Fax : 05 4000 21 23
> http://www.math.u-bordeaux.fr/maths/cellule
>
> _______________________________________________
> torqueusers mailing list
> torqueusers at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torqueusers
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/torqueusers/attachments/20061124/5fb3c77c/attachment-0001.html


More information about the torqueusers mailing list