qsub_string-ref to specify number of nodes and ppn on PBS cluster

Submitted by anans on Wed, 2016-08-03 10:40

Forums

Frealign

Dear All,

I've been trying to modify the qsub_string_ref of mparameters to specify a queue and the number of nodes and ppn for our PBS cluster. I'm using the following:

qsub_string_ref "-q mpi -l nodes=3:ppn=64"

It's recognising the -q command and sending the job to our mpi queue but it's ignoring the -l nodes=3:ppn=64 and sends the job to other nodes. Does anyone have a successfully working qsub_string_ref for PBS clusters?

Thanks,

Andrea

niko Wed, 2016-08-03 12:23

Please note that Frealign does not use MPI (just in case you were trying to send the jobs to the mpi queue to parallelize using MPI). Regarding the other options, are they honored when you submit a job other than Frealign on your cluster?

anans Wed, 2016-08-03 12:41

Hi Niko,

Yes, those options are recognized and honoured when we submit other jobs to our cluster. Thanks for the reminder about not using MPI. We have just one queue and it's named MPI but anything can be run through this queue. I am just trying to get frealign to avoid going to certain nodes so I thought the -l nodes=x:ppn=x would do the trick.

Thanks,

Andrea

niko Wed, 2016-08-03 12:48

It is surprising that the options work when used with other jobs. Is there a way to get the submission string that is send to the queue so confirm that all options are properly transmitted? This could probably be done with some queue list command (I am sorry but I am not that familiar with the PBS system). If the options do not work, maybe one could create a Frealign queue that only includes the nodes that are supposed to run Frealign?

anans Thu, 2016-08-04 06:24

Looks like the mult-refine.com script sends X number of jobs (based on the entry in nprocessor_ref) that each request 1 node and 1 task (-l nodes=1:ppn=1). So that's why it was ignoring my own entry in the qsub string ref. The easiest thing will be to create a second queue to exclude frealign from certain nodes. Thanks for the advice!

Cheers,

Andrea