-l procs = X is an unknown node-attribute

[jdmoore2004]

See also #207 which reports this issue. I am reposting as a new issue, since it has been open for more than 3 years and still seems to be an issue.

Using the 6.1.0 I am also finding.

#PBS -l procs=X

yields "Not Running - PBS Error: Unknown node-attribute"

I am using pbs_sched

I have removed all resources_default.ncpus, resources_default.nodect, and
resources_default.nodes from my queues.

I can run across multiple nodes, using #PBS -l nodes=X:ppn:Y just fine.

From this post several years ago

http://www.supercluster.org/pipermail/torqueusers/2013-December/016595.html

And the latest user manual:

http://docs.adaptivecomputing.com/torque/6-1-0/adminGuide/help.htm#topics/torque/2-jobs/requestingRes.htm?Highlight=ppn

I thought #PBS -l procs=X

should work with pbs_sched and pbs_sched should be able to schedule the
jobs across the nodes on the proper number of nodes?

Does anyone know if this works with the latest version of Torque? It seems like since it isn't being passed to the scheduler it wouldn't work for Moab or Maui either.

[Glen Beane]

It works with Moab. The value is recorded in the job's Resource_List, which Moab is able to parse. Moab then passes an exec host list back to pbs_server with the proper number of procs when it runs the job.

This was actually fixed in Torque 2.5.0 (see change log entry below), but it has been broken for at least 3 years now.

e - Enabled TORQUE to be able to parse the -l procs=x node spec. Previously
TORQUE simply recored the value of x for procs in Resources_List. It
now takes that value and allocates x processors packed on any available
node. (Ken Nielson Adaptive Computing. June 17, 2010)

[jdmoore2004]

I downgraded to 2.5.6 and "-l procs" works with pbs_sched and qrun. It actually seemed to work with qrun in 6.1.0.

2.5.6 seems to work for me, so I'll stick with that until/if they fix pbs_sched.