Yes.. it starts new shells on each machine, and doesn't necessarily take all the env vars with it. I suggest setting them in your script which you feed to qsub, although that is also only run on a single compute node. If you want environment propagated to every node in your job, it gets worse-
you have to make a script which is called by your parallel launcher (pbsdsh, mpirun, or whatever you choose), which is called by your qsub target script, which is called by qsub. Hope you followed that- it's ugly, but the reality.

Jeremy


At 11:27 AM 1/9/2003 -0500, Brian Williams wrote:
Well I have made some progress and got a few simple things to run. I do have
one curious problem.
In my normal users' environment I have some $variables for jave, like
$CLASSPATH, that I see fine on my env. but when I submit my request to pbs,
it fails and says CLASSPATH: Undefined variable. at the point where it
should start the java. the rest of the script seems to work fine at that
point. So, I have $classpath in my env, but when i run the script you have
here, i don't see it and a few others. Any idea why these variables aren't
being propagated during the PBS run?
thanks,
Brian

-----Original Message-----
From: Jeremy Enos [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, January 07, 2003 10:49 AM
To: Jeff Squyres; Brian Williams
Cc: 'Oscar'
Subject: Re: [Oscar-users] Running a single job on multiple machines


Even more handy is the $PBS_NODENUM env var.  qsub a script which contains
/usr/bin/env to get the full job environment available to you.  For
starters, do something like

#!/bin/bash
echo
echo My hostname is $HOSTNAME
echo My environment:
/usr/bin/env
echo

         Jeremy

At 07:57 AM 1/7/2003 -0500, Jeff Squyres wrote:
>On Mon, 6 Jan 2003, Brian Williams wrote:
>
> > [snipped]
> > Or, to schedule a series of these script runs with the numbers already
> > configured and to just schedule them out on the cluster. Should I use
> > PBS or mpirun or something different? The documentation is a little
> > overwhelming at first, especially for PBS.
>
>Yes, the docs for PBS are somewhat complex, but definitely worth the
>effort to understand.
>
>PBS is probably your best bet -- some creative use of pbsnodes and the
>$PBS_NODEFILE environment variable may give you what you describe.  For
>example:
>
>- Number of free nodes currently under PBS control (from any node):
>         pbsnodes -a | grep "state = free" | wc -l
>- Number of nodes in a PBS job (from the within the job itself):
>         wc -l < $PBS_NODEFILE
>
>Hope that helps.
>
>--
>{+} Jeff Squyres
>{+} [EMAIL PROTECTED]
>{+} http://www.lam-mpi.org/
>
>
>-------------------------------------------------------
>This SF.NET email is sponsored by:
>SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
>http://www.vasoftware.com
>_______________________________________________
>Oscar-users mailing list
>[EMAIL PROTECTED]
>https://lists.sourceforge.net/lists/listinfo/oscar-users


-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com
_______________________________________________
Oscar-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-users

Reply via email to