Hi Nele Do you run that in parallel or on a single node? In any case you need enough memory to be able to process HCP data. Bedpostx calls fsl_sub to send jobs to the cluster queue. If your cluster’s default memory allocation limits are low, you can easily run out of memory.
You could slightly modify the bedpostx script by adding to the fsl_sub calls the -R option that determines the amount of memory you request from a node. Hope that helps Stam On 4 Sep 2014, at 14:48, Nele De Witte <[email protected]> wrote: > Dear HCP experts, > > We've been having difficulties with running BEDPOSTX on the cluster and > we do not know what the problem is, so if anybody would have any > suggestions, we would be very grateful. The problem is the following. > > When running the following command on the clusters with FSL 5.6 > > bedpostx ./DTI/s1 -n 3 -b 1000 -model 2 -w 1 -g > > It arrives to process a certain number of slices (17 last time), then it > just stops. > The thing is that it does not stop with an error message, or with > something that IT people can identify. It just stops. > We don't know if it is a problem with a certain FSL-version (we tried > older versions, but they also didn't work) of with the cluster.... > > > We also tried running bedpostx on a normal computer with the latest FSL > version. We ran the following command: > > bedpostx dti_folder_path -n 3 --model=2 > > and it worked. We guess the command is the one as above, since -b 1000 > and -w 1 are default values. On the other hand -g was not recognized as > the correct option on the desktop, and when we added it to the command, > it did not work. But in this case it was not the same problem we had on > cluster (stopping after a number of slices), it just stopped. > To us, this looks like a mixed problem between a FSL one (version, > commands, etc), and one connected to the architecture (cluster vs pc).... > > Does anybody have any thoughts on how we could solve this? > > Thanks! > > Best, > > Nele De Witte > Psychophathology and Affective Neuroscience lab (PANlab) > Ghent University > > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
