[GE users] parallel jobs submission help

Wheeler, Dr M.D. mdw10 at leicester.ac.uk
Thu Nov 25 21:44:19 GMT 2004


    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

I have possibly found the problem if is qstat -j i get
 
scheduling info:            queue "compute-0-1.q" dropped because it is temporarily not available
                            queue "compute-0-2.q" dropped because it is temporarily not available


 

________________________________

From: Reuti [mailto:reuti at staff.uni-marburg.de]
Sent: Thu 25/11/2004 21:28
To: users at gridengine.sunsource.net
Subject: RE: [GE users] parallel jobs submission help



Quoting "Wheeler, Dr M.D." <mdw10 at leicester.ac.uk>:

> OK I think that I have gone some way to sorting my problem out
> 
> I have 3 compute nodes set up on my system, however sge only appears to
> submit jobs to the first node
> 
> if i run a script with the line
> 
> > qsub -pe mpich 2 molpro.sh
> 
> where molpro.sh is
> 
> #!/bin/bash
> #
> #$ -cwd
> #$ -j y
> #$ -S /bin/bash
> molpro -n2 test.com
> 
> everything runs ok (on two procs)
> 
> now if I submit another job
> 
> > qsub -pe mpich 2 molpro1.sh
> 
> where molpro1.sh
> 
> #!/bin/bash
> #
> #$ -cwd
> #$ -j y
> #$ -S /bin/bash
> #molpro -n2 test1.com
> 
> this second job sits in the queue
> 
> do I have to tell it which queue to allocate the job to???

No, SGE will select the appropriate ones. I guess the "#molpro" is a typo.

When you set: "shell_start_mode     unix_behavior" in the queue definition, you
don't need the -S switch.

What is your PE definition ("qconf -sp mpich"), are there enough slots given
(the slots there are slots in total inside this PE, so should be six in your
case I think).

And you need this lines in your script:

export PBS_NODEFILE=$TMPDIR/machines
export SCRATCH=$TMPDIR
molprop -n $NSLOTS $1

Then you can use this script for all combinations, and all files should go to
the scatch directory on the node - not to the root level of /tmp. This way SGE
can make a proper cleanup after the job. $NSLOTS will be replaced with the
number of slots you got. Maybe Molpro is doing the right thing anyway, because
you are just on one node for now.

Then you can submit with:

qsub -pe mpich 2 molpro.sh test.com

CU - Reuti

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net






    [ Part 2: "Attached Text" ]

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net



More information about the gridengine-users mailing list