Running FDS on a LSF Linux Cluster

412 views
Skip to first unread message

steve...@gmail.com

unread,
May 19, 2015, 7:11:00 AM5/19/15
to fds...@googlegroups.com
Hi everyone,
I would like to submit an FDS simulations on a LSF Linux cluster 

I am struggling to work out the right syntax for the .sh file I will need to submit the job
Is anyone willing to share an example I could use as starting point?

Many thanks
Ste

Glenn Forney

unread,
May 19, 2015, 10:19:03 AM5/19/15
to fds...@googlegroups.com
does your cluster have a queuing system (like PBS, Torque etc)?

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/0de93ca2-a817-4d5f-8635-e11cb8934ca1%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.



--
Glenn Forney

steve...@gmail.com

unread,
May 19, 2015, 1:03:38 PM5/19/15
to fds...@googlegroups.com
Yes it does
Torque 

thanks
ste

Glenn Forney

unread,
May 19, 2015, 3:29:12 PM5/19/15
to fds...@googlegroups.com
if you have the FDS-SMV repository on your Linux cluster, you could use the qfds.sh script in Utilities/Scripts to submit a job to a torque cluster.  qfds.sh generates a script that Torque uses to submit a job with qsub .  For example, the command

 qfds.sh  -p 8 -e ~/FDS/FDS6/bin/fds mplume5c8.fds

generates the following qsub script (which you run using qsub script_name where script_name is a file containing the script below).  You could adapt the following script for your cluster or use the above qfds.sh command.  Note the line containing /home/gforney/FDS/FDS6/bin/fds mplume5c8.fds appears on the same line as mpirun

#!/bin/bash
#PBS -N mplume5c8(MPI_IB)
#PBS -e /home/gforney/FDS-SMV/Verification/Visualization/mplume5c8.err
#PBS -o /home/gforney/FDS-SMV/Verification/Visualization/mplume5c8.log
#PBS -l nodes=4:ppn=2
#PBS
export OMP_NUM_THREADS=1

cd /home/gforney/FDS-SMV/Verification/Visualization
echo Start time: `date`
echo Running mplume5c8 on `hostname`
echo Directory: `pwd`
/shared/openmpi_64ib/bin/mpirun --report-bindings --bind-to socket --map-by socket -np 8 /home/gforney/FDS/FDS6/bin/fds mplume5c8.fds




For more options, visit https://groups.google.com/d/optout.



--
Glenn Forney

steve...@gmail.com

unread,
May 21, 2015, 6:09:55 AM5/21/15
to fds...@googlegroups.com
I will give ti a shot 
THanks Gleen

Ste

Stefano Capra

unread,
Oct 8, 2015, 10:23:55 AM10/8/15
to FDS and Smokeview Discussions
Hi Glenn,
I have been trying to test the qsub script running:
bash qfds.sh -p 48 -e /gpfs/stfc/local/apps/intel/FDS/6.2.0/bin/fds test.fds 

It generates the following output:

         Input file:test.fds
         Executable:/gpfs/stfc/local/apps/intel/FDS/6.2.0/bin/fds
              Queue:batch
              Nodes:24
          Processes:48
 Processes per node:2


But it is not writing down a qsub script 
Is there any way to set the number of cores per node?
Do I have to split the mesh beforehand in the same number of subdivision of the number of cores I am using?

Thanks
ste

Randy McDermott

unread,
Oct 8, 2015, 10:30:55 AM10/8/15
to FDS and Smokeview Discussions
We use -n [number of cores per node].  And, yes, you must have at least one mesh per core.

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.

Stefano Capra

unread,
Oct 8, 2015, 10:57:38 AM10/8/15
to FDS and Smokeview Discussions
Thanks Randy!
Should it write down the script before submitting it?

Thanks
ste


On Tuesday, May 19, 2015 at 12:11:00 PM UTC+1, Stefano Capra wrote:

Randy McDermott

unread,
Oct 8, 2015, 11:07:49 AM10/8/15
to FDS and Smokeview Discussions
Not sure I understand the question.  If you do this (note the -v at the end):

bash qfds.sh -p 48 -n 2 -e /gpfs/stfc/local/apps/intel/FDS/6.2.0/bin/fds test.fds -v

you should see the details of the script without actually submitting it.

I've never run qfds.sh outside our cluster, so I am not sure how that works.  Unfortunately, both Kevin and Glenn on out today.  Kevin is back tomorrow I think and may have better advice.


--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.

Glenn Forney

unread,
Oct 8, 2015, 11:27:01 AM10/8/15
to fds...@googlegroups.com

The qfds script erases the qsub script after it runs it. You could edit qfds.sh to change that behavior.

Stefano Capra

unread,
Oct 10, 2015, 4:46:31 AM10/10/15
to FDS and Smokeview Discussions
Hi Glenn and Randy,
the -v flag at the end didn't help, the script is still not running and not showing me the submission script 
I will look through the code today and try to modify it for my needs

thanks
ste


On Tuesday, May 19, 2015 at 12:11:00 PM UTC+1, Stefano Capra wrote:

Glenn Forney

unread,
Oct 10, 2015, 9:41:43 AM10/10/15
to fds...@googlegroups.com

The -v flag outputs script generated to stdout. Redirect output to a file, make any edits you want, then use qsub to run it.

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.

Stefano Capra

unread,
Oct 14, 2015, 12:15:19 PM10/14/15
to FDS and Smokeview Discussions
Hi everyone,
I sorted the problem writing a small script for the submission file generation.
I turned out to be quite different since I am running it on an LSF system

Thanks
ste


On Tuesday, May 19, 2015 at 12:11:00 PM UTC+1, Stefano Capra wrote:
Reply all
Reply to author
Forward
0 new messages