[Octopus-users] octopus on the grid

Emily listemily at eml.cc
Fri Jul 11 22:37:52 WEST 2008


Hi Goranka,

I use octopus on a parallel cluster that uses qsub and .pbs files to
submit jobs to a queue, which then assigns them to a node (processor). 
So my job can run on any of several computers, and I don't know which
one it will be.  However all of them do have access to a common
filesystem.  But I copy relevant files on to the particular node so that
we don't have to spend time writing to the common filesystem until the
end of the job.  If that is a similar situation to yours, perhaps this
will be helpful.  Here is a little snip from my jobsubmit.pbs file:

    ssh -n $nodeFree mkdir -p ${TMPDIR}/${jobDir}
    scp ${jobDir}/${INFILE} ${nodeFree}:${TMPDIR}/${jobDir}
    ssh -n $nodeFree ln -s ${TMPDIR}/${JOB} ${TMPDIR}/${jobDir}/${JOB}
    ssh -n $nodeFree "( cd ${TMPDIR}/${jobDir}; \
        ./${JOB} > ./${LOG}; \
        scp -r ./* $PBS_O_WORKDIR/${jobDir} )"&
    echo "Submitted ${jobDir} to $nodeFree on `date`."

We start out in some working directory on the common filesystem, from
which we ran the jobsubmit.pbs, $PBS_O_WORKDIR.  It contains several
subdirectories for jobs that I want to submit.  Typically I have 8
directories, each containing an inp file with a different value for one
variable.  These 8 directories are the ${jobDir} directories.

There is a list somewhere of free processors, and somehow we've been
assigned $nodeFree to work on.  So we make a directory there.

We copy the inp file to that directory.

We make a symbolic link from the octopus executable in the common
filesystem to that directory, so we can run octopus from that directory.

We cd into that directory, run octopus, redirecting output to a log
file.

We copy recursively everything that ends up in that directory back to
the relevant job directory in the working directory on the common
filesystem.

We tell the user we submitted the job in that particular directory.  

I'll attach the whole file in case it is useful to you.  

Good luck,
Emily

----- Original message -----
From: "Goranka Bilalbegovic" <gbilalbegovic at gmail.com>
To: octopus-users at tddft.org
Date: Fri, 11 Jul 2008 19:22:29 +0200
Subject: [Octopus-users] octopus on the grid

It looks as if the Octopus users mailing list does not work now, but ...

I think that my problem with the octopus on a grid is that my job
actually runs
not in /home/gbilalbe/ but in /home/gbilalbe/gram_scratch*/
Then this scratch directory is immediately deleted when job finishes and
I am
not able to see what is there. I think that the octopus directories
exec, restart and static are formed in this gram_scratch.

My question now is what is the best way to run octopus under given
conditions ?

Thanks for any help.
Best regards,
Goranka
_______________________________________________
Octopus-users mailing list
Octopus-users at tddft.org
http://www.tddft.org/mailman/listinfo/octopus-users
-------------- next part --------------
A non-text attachment was scrubbed...
Name: jobsubmit.pbs
Type: application/octet-stream
Size: 7510 bytes
Desc: not available
Url : http://www.tddft.org/pipermail/octopus-users/attachments/20080711/61b72928/attachment.obj 


More information about the Octopus-users mailing list