[torqueusers] PBS nodes problem
gwimpey at mines.edu
Tue Oct 19 12:05:55 MDT 2004
If I recall, Maui's interpretation of nodes/ppn is "weak". That is, it
will just multiply them together and allocate the job across whatever
CPUs are available. Or have I misunderstood the docs? Of course, this
only applies if you are using Maui as a scheduler.
I would assume that the nodes/ppn/ncpus syntax is a historical legacy of
PBS. One has to be careful about changing these things so as not to
break existing installations.
On Mon, 2004-10-18 at 14:57, Curtis W. Hillegas wrote:
> The -lnodes=#:ppn=# syntax works very well for dedicated clusters where
> jobs can be specifically allocated to maximize performance, etc. But in
> shared clusters this causes confusion, especially for newer users. Many
> users in a shared cluster would rather have their job start earlier with
> a mix of one, two, or however many processors per node rather than wait
> until there is an ideal block of free nodes available. Of course the
> scheduler can be configured to fill in the processors more the way users
> would like it, but the concept is sometimes confusing to the end user.
> Why was the choice made not to allow the user to say "I want n
> processors allocated however you can fit them on the nodes that
> currently have unallocated processors?"
> Danny Sternkopf wrote:
> > Hi,
> > ncpus should be used for big SMP machines only.
> > It specifies the number of cpus for one single node.
> > If you have one or two processors machines in
> > cluster then you have to specify the number of nodes.
> > --> 'lnodes:<number of nodes>:<number of cpus per node>
> > For example: -lnodes=15:ppn=2 will request 30 CPUs
> > on Dual machines.
> > Best regards,
> > Danny
> > On Sun, Oct 17, 2004 at 01:39:19AM +0800, torqueusers at supercluster.org wrote:
> >>Is't compulsory to put "#PBS -l nodes" in the job script? Can I just
> >>use "#PBS -l ncpus" in my script to submit job? My script failed to
> >>work when I submit without declared no. of nodes? Pls help... Thank
> >>torqueusers mailing list
> >>torqueusers at supercluster.org
> > _______________________________________________
> > torqueusers mailing list
> > torqueusers at supercluster.org
> > http://supercluster.org/mailman/listinfo/torqueusers
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers