[torqueusers] jobs don't run if server_priv/nodes includes master

Isaac Hailperin ihailper at googlemail.com
Mon Mar 30 06:06:01 MDT 2009


I found that if I include the cluster headnode (which is also the torque
submit host) in the file /var/spool/torque/server_priv/nodes, I need to
request one more node then nessessary for satisfying my -np #procs
That is I need to specify nodes=3:ppn=8 to be able to request -np 16.
Or nodes=4:ppn=8 to be able to request -np 24.

Why is that? I can see that for large clusters it does not make sense to
have batch jobs running on the headnode. But for small clusters, where
the master is not doing much, it would be a waist of ressources to not
include it. So is it just a "missing feature"?

Btw, I am using torque 2.3.6 on rhel 5.3.


More information about the torqueusers mailing list