[torqueusers] File descriptor limits with mpiexec and pbs/torque

Dave Jackson jacksond at supercluster.org
Tue Nov 16 20:28:03 MST 2004


Chris,

  Thanks for the info.  This has been added to the 'Scalability
Guidelines' overview of section 3.1 in the TORQUE admin manual - see

http://clusterresources.com/torquedocs/3.1communication.shtml

 If anyone else has suggestions for large systems, large jobs, or large
numbers of jobs, please let us know.

Dave
Cluster Resources, Inc

On Wed, 2004-11-03 at 13:59, Maestas, Christopher Daniel wrote:
> Hello,
> 
> We recently discovered when launching > 504 processes on a system you
> hit a file descriptor limit.
> If you add the following to the startup script of your pbs_mom:
> ----
> ulimit -n 4096
> ----
> 
> And ensure that the fs.file-max parameter is high enough in
> sysctl.conf
> ---
> fs.file-max = 65536
> ---
> 
> You can launch job sizes > 504.  This is tested with torque 1.1.0p4
> and mpiexec-0.77 ….
> We modified the pbs_mom startup script from a modified Nico rpm for
> 1.1.0p4. :-)
> 
> -- Chris
> 
> 
> 
> ______________________________________________________________________
> _______________________________________________
> torqueusers mailing list
> torqueusers at supercluster.org
> http://supercluster.org/mailman/listinfo/torqueusers



More information about the torqueusers mailing list