[torqueusers] forcing use of one processor per node?

Troy Baer troy at osc.edu
Wed Apr 16 12:00:39 MDT 2008


On Wed, 2008-04-16 at 13:29 -0400, John Young wrote:
> Many thanks to all who responded!
> 
> To answer some questions, yes, I am using maui with torque.
> 
> The job I am running does use HP-MPI, and I was not sure
> whether Torque was handing the application a bad mp_host_list
> or whether the application was misbehaving.  Based one one
> of the suggestions received, I added to my PBS batch file a
> command to list out the contents of $PBS_NODEFILE
> 
> #!/bin/sh
> #PBS -l nodes=4:ppn=1
> #PBS -q default 
> #PBS -j oe 
> #PBS -N s1.PBS 
> echo 'This is the list of nodes that Torque has assigned to this job:'
> cat $PBS_NODEFILE
> .
> .
> .
> 
> And despite specifically asking for 'ppn=1', I got:
> 
> Warning: no access to tty (Bad file descriptor).
> Thus no job control in this shell.
> This is the list of nodes that Torque has assigned to this job:
> n16
> n16
> n17
> n17
> 
> So it does not appear to me that the application is at fault.  :-/

I suspect this is coming from your scheduler configuration rather than
TORQUE.  IIRC, the defaults in Maui allow it to rewrite the node
specification so long as it results in the requested number of
processors.  If you want it to follow your node specification exactly,
you may need to set the following in your maui.cfg:

JOBNODEMATCHPOLICY EXACTNODE

	--Troy
-- 
Troy Baer, Supercomputing Support Engineer
Ohio Supercomputer Center
Phone:  614-292-9701
http://www.osc.edu/supercomputing/



More information about the torqueusers mailing list