[torqueusers] policy not working as expected?

Brandon Sawyers brandor5 at gmail.com
Wed Mar 7 08:34:46 MST 2012


Hello everyone:

I posted this to mauiusers a couple of days ago but have seen no reply.

We are bringing up a new system and are running into an issue with maui.

We want jobs to behave like this. A user requests a number of nodes
regardless of ppn and gets that number of nodes (nodes=6:ppn=1).  At the
same time, we want only one job to be running on a node at one time. So
that user would get 6 nodes and no other jobs would be able to run on those
nodes while that job is running. Even though that user is only using 1 of
the available cores.

We expected the following two config changes to make this happen.

JOBNODEMATCHPOLICY    EXACTNODE
NODEACCESSPOLICY      SINGLEJOB

While only one job will run on a node like we want, but (using the example
above) all 6 cores of that node are getting used, instead of using 1 core
on 6 different nodes.

Interestingly, the following nodes=1:ppn=1 gives me 1 core from 1 node.
nodes=1:ppn=(2-6) gives me 6 cores.

Something to be aware of, when I say node, I mean numa node.

What are we missing?

Thanks,
Brandon
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/torqueusers/attachments/20120307/f53dbb18/attachment.html 


More information about the torqueusers mailing list