Ok, I have as part of my config:

NODEALLOCATIONPOLICY            MINRESOURCE

In my cluster I have mostly 8 processor nodes.
I have one node that has 16 processors and 2 gpu cards. They are in the nodes 
file as such:

tesla np=16 gpus=2 tesla

So someone could:
qsub -l nodes=1,tesla

now I expected that with MINRESOURCE as my NODEALLOCATIONPOLICY, any job with 
#PBS -l nodes=1:ppn=8 would go to one of the more 'normal' 8 processor nodes 
before hitting the tesla node.
This does not seem to be the case as I have completely free 8 processor nodes, 
but non-gpu/tesla jobs are being run on the tesla node, blocking any jobs that 
need the gpus.

Is there something I am missing about the NODEALLOCATIONPOLICY? Am I mistaken 
in believing it would go to non-gpu systems before the gpu system? From the 
documentation:

"Those nodes with the fewest configured resources which still meet the job's 
resource constraints are selected. "



Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238



Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238

_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers

Reply via email to