[torquedev] Question reg. PBS_NODEFILE

[Brock Palen]

You can,  as long as you allocate per cpu in nodes.  We do this  
because we have a mix of 2 and 4 cpu boxes.  If i ask for

-l nodes=4:ppn=1

I get PBS_NODEFILES that look like

host1
host1
host1
host1

or

host2
host2
host1
host1

even

host1
host2
host3
host4

Sorry to drag that out :-)
I believe this is schedular/config dependent. We do it with moab and  
maui.

Brock Palen
Center for Advanced Computing
brockp at umich.edu
(734)936-1985


On Nov 30, 2006, at 2:55 PM, Glen Beane wrote:

> On 11/30/06, Garrick Staples <garrick at clusterresources.com> wrote:
>> On Thu, Nov 30, 2006 at 07:51:15AM +0100, ?ke Sandgren alleged:
>> > On Wed, 2006-11-29 at 12:03 -0700, Garrick Staples wrote:
>> > > On Wed, Nov 29, 2006 at 01:41:18PM +0100, ?ke Sandgren alleged:
>> > > > Hi!
>> > > >
>> > > > PBS_NODEFILE isn't gettting set unless -l nodes= is specified.
>> > > > I.e. only when neednodes is set, -l ncpus=x doesn't give a  
>> nodefile.
>> > > >
>> > > > Intentional or "bug"?
>> > >
>> > > Yes, if the job isn't assigned "nodes", then you don't get a  
>> nodefile.
>> >
>> > Yes, but it will get assigned some nodes (or hosts if you prefer  
>> that)
>> > or it wouldn't run at all so why not always set this?
>>
>> "nodes", meaning, the job attribute.
>>
>> The whole concept of "ncpus" in OpenPBS/TORQUE is buggy and crufty.
>
> I agree there.  It would be nice if there were a clean and reliable
> way to say "i don't care how many nodes you give me as long as I have
> N cpus", and still get the appropriate PBS_NODE_FILE generated, etc
> _______________________________________________
> torquedev mailing list
> torquedev at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torquedev
>
>