The fsl_sub command uses SGE to submit these commands as one parallel
array, i.e., to port to a non-sge cluster you need to replicate the
qsub command used. Your admin needs to port all the functionality in
fsl_sub. If you can give details of your cluster implementation then I
might be able to suggest the correct command.
Dave Flitney, IT Manager
University of Oxford, FMRIB Centre
JR Hospital, Oxford OX3 9DU
T: 01865 222713 F: 01865 222717
On 2 Jul 2009, at 02:04, Michael Scheel <[log in to unmask]>
wrote:
> Dear FSL-List,
>
> I have the opportunity to use a Cluster for my BedpostX Analysis to
> speed up the processing time.
> With a little (actually quite a big) help from the admin I was
> successful setting up fsl on the cluster.
>
> One question though from the admin that I want to forward here:
>
>> In the output directory, diffusion.bedpostX, there is a file called
>> commands.txt containing:
>> /global/home/phillips/software/fsl/fsl414/bin/
>> bedpostx_single_slice.sh /global/home/phillips/software/fsl/run/
>> diffusion 2 1 1000 125 0 25 0
>> /global/home/phillips/software/fsl/fsl414/bin/
>> bedpostx_single_slice.sh /global/home/phillips/software/fsl/run/
>> diffusion 2 1 1000 125 0 25 1
>> ...
>> continuing from 0 to 56.
>> Are all these commands independent? Could you have set up this
>> job so that you call bedpostx_single_slice.sh yourself instead of
>> getting bedpostX to do it for you?
>> In that case, the whole thing could have been done as 56 separate
>> jobs - that would have been finished in an hour and a half.
>
> I know that the suggested sge_cluster environment handles splitting
> up the jobs for every single slice by just running the bedpostx
> command.
> I suppose that this is handled at the beginning of the bedpostx
> program
>
> if [ "x$SGE_ROOT" = "x" ] ; then
> if [ -f /usr/local/share/sge/default/common/settings.sh ] ; then
> . /usr/local/share/sge/default/common/settings.sh
> elif [ -f /usr/local/sge/default/common/settings.sh ] ; then
> . /usr/local/sge/default/common/settings.sh
> fi
> fi
>
> In our case we don't use sge to run the cluster. Would it then be
> reasonable to use <fslsplit data.nii.gz diff_slices/data_slice_ -z>
> and then run bedpostx_single_slice.sh ?
> Any word of caution before we try that?
>
> Thanks, Michael
>
|