I've attached our workaround for this. This version of fsl_sub (which
bedpostx calls) works with PBS.
Peace,
Matt.
On 8/22/13 5:35 PM, "Jenifer Juranek" <[log in to unmask]> wrote:
>Is anyone familiar with PBS cluster environment? We are working on a
>mechanism for sending our bedpostx jobs to our cluster; however, it turns
>out we have PBS instead of SGE.
>
>Although we could edit the fsl_sub to make it work for a PBS system, this
>would violate rules of the queue mechanism set up on our PBS cluster. On
>our PBS cluster, the user has to run everything within a job script.
>Running a master script like bedpostx which spawns a few qsub jobs (by
>fsl_sub) is not allowed.
>
>We're getting some advice from our "cluster guy" who is not that familiar
>with FSL; although we know FSL on local machines, we are not yet familiar
>with sending jobs to our PBS cluster. Hopefully someone on the list can
>provide some advice?
>
>Our cluster guy was thinking a solution might be be to break up the
>majority of bedpostx into three PBS style job scripts: 1) preprocessing
>(bedpostx_preproc.sh) 2) real processing (with commands.txt) and 3) post
>processing (bedpostx_postproc.sh). I would assume the processing with
>commands.txt is the most time consuming step right? If that's the case,
>it may be easier to run everything else locally and run the tons of
>bedpostx_single_slice.sh commands on the PBS cluster in parallel.
>
>Any thoughts?
>
>Many Thanks,
>Jenifer
|