[torqueusers] torque only assigns one node

Garrick Staples garrick at usc.edu
Sun Feb 10 22:23:58 MST 2008


On Sat, Feb 09, 2008 at 03:27:12AM -0500, asubedi alleged:
> I am trying to use "#PBS -l nodes=3" in my job script to run a job in
> three nodes. However, torque only assigns one node. However, If I do
> "#PBS -l nodes=node001+node002+node003", it assigns those three nodes.
> Does anyone know what am I doing wrong? The output of qmgr -c 'p s'
> is:

You have too many resources_min/max/default set.  It is confusing the scheduler.

unset queue long resources_max.cput
unset queue long resources_max.ncpus
unset queue long resources_max.nodes
unset queue long resources_min.cput

unset server resources_available.nodect
unset server resources_default.cput
unset server resources_default.nodect

And restart pbs_server.


"nodes" is a string, min/max.nodes is meaningless.
"nodect" is an integer computed from nodes, default.nodect is meaningless.

"cput" is largely unreliable unless you can enforce that all processes are
children of pbs_mom.  "walltime" works well for all cases.

"ncpus" can't be used with "nodes" because they are contradictory.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20080210/c81322f8/attachment.bin


More information about the torqueusers mailing list