[torqueusers] a question about $usecp

Joshua Bernstein jbernstein at penguincomputing.com
Thu Jul 8 11:48:47 MDT 2010


Hell Andreas,

$usecp needs to be enabled when you have your current working directory 
(CWD) or really the directory you qsub'd from NFS (or AFS etc) mounted 
on the compute nodes. Otherwise the compute nodes attempt to SCP (or in 
builds without --with-scp, RCP) the .o and .e files.

For example, on basic clusters /home is NFS mounted on the compute nodes 
as well as the submission nodes. So I need a line in mom_priv/config 
that tells pbs_mom to simply "use cp" rather then rcp/scp

$usercp	*:/home /home

If I then qsub a job out of /home/jbernstein, the output files get 
copied using 'cp'.

If your jobs aren't even starting, then *.o and *.e files aren't being 
created and there is another problem afoot.

-Josh

Andreas Davour wrote:
> I have been having troubles with getting my jobs to start (mentioned in 
> another letter), and have started to look at the $usecp directive.
> 
> Reading the Torque Admin Guide I still don't understand how it's supposed to 
> work.
> 
> Is it only for copying the final e# and o# files? My problem seem to end up 
> earlier (I guess they never start), since I see no such files anywhere, not 
> even in the undeliverable folder. 
> 
> I have my $HOME in afs, both on the worker node and on the submit node.
> 
> /andreas


More information about the torqueusers mailing list