Hi,
I am running a large tbss analysis containing 825 individuals, and have
some problems with tbss_2_reg -n. We are running the analysis on a linux
cluster with a pbs (public batch system), and tbss_2_reg runs in
parallell just fine when I have used it for 150 individuals.
Our problem is that fsl uses a really long time only using one cpu (and
not the 800 cpus that are avaliable for my analysis), and I think that
it uses all this time working with imglob. And the process using imglob
apparently is not running in parallell.
The next thing is that fsl uses a long time making the
pbsdispatch_commands file.
So it seems that fsl is using a long time getting to the fsl_sub part
where the process is running in parallell. Have I got this right, or am
I missing something?
My questions are as follows:
1. Is it possible to also run the imglob in parallell?
2. Or is it possible to first run imglob on one cpu as an individual
job, and then start the parallell jobs?
3. Or can we split the analysis in smaller parts, for example compare
100 images in one analysis and then do several analysis and finally
bring them all together when continuing with tbss_3?
Thank you in advance!
Best regards,
Live Eikenes
Post doctor
Trondheim fMRI group
Norwegian University of Science and Technology
Trondheim, Norway
|