[torqueusers] Re: [Mauiusers] MAXNODES broken?

Dave Jackson jacksond at supercluster.org
Wed Nov 3 16:41:05 MST 2004


Chris,

  We have already started looking at modifying TORQUE to allow
specification of whether 'nodes' bounds checking is constrained by nodes
available or by procs available.  Sorry, you have been on our list for a
while, we've just been too busy to respond.  Expect a fix within 3
weeks.

Dave

On Wed, 2004-11-03 at 15:04, Chris Samuel wrote:
> On Thu, 4 Nov 2004 08:05 am, John Wagner wrote:
> 
> > No easy way to get a random allocation. But if you are using a cluster
> > built with bi-proc nodes then
> >    qsub -l nodes=45:ppn=2
> > will give you 90 processors spread over 45 nodes (i.e 45 nodes with 2
> > job processors each on them).
> > Thus
> >    qsub -l nodes=45:ppn=2+1:ppn=1
> > will give you  45 nodes with 2 job processors allocated plus one node
> > with 1 job processor allocated giving you 91 processors in total.
> 
> Yup, I'm aware of this, but it doesn't help when you've got 90 CPUs spread 
> around in a fairly haphazard manner due to other folks jobs.
> 
> What we really want is a way to say to someone wanting to run a large job 
> "just do this and you'll get N (where N>90) CPUs across the cluster", 
> otherwise you've got to wait a long time (potentially months with some of our 
> users) to get your job running for your requested configuration to appear..
> 
> > I don't know of anyway to say you want 91 processors randomly
> > distributed anywhere there are free "job slots" within the cluster.
> 
> This is what I'm really after, anyone got any clues ?
> 
> > I hope this helps to solve your problem.
> 
> Sadly not, this has been a long standing whinge of mine.. :-)
> 
> Chris



More information about the mauiusers mailing list