For what its worth, on our workstations we run each subject's anatomy pipeline as a separate thread, which takes 8-12 hours, then use the parallelization to speed up the fMRI pipelines with either SGE/OGE or ht_condor, with a thread devoted to each bold run being processed
__________________________ gaurav patel [email protected] www.neurofreak.net On Apr 27, 2017, at 9:44 AM, Harms, Michael wrote: > > The number of windows you use to launch the processing is irrelevant. But if > you only have 8 cores on your machine, and only want to use 7 of them, you > can’t run 50 subjects at once. > > In this case, where you have a bunch of subjects to process, I would run 7 > subjects at a time, while limiting each to just 1 core, by setting whatever > variable in the pipeline that is used to limit/control the number of threads > used for multi-thread capable binaries. > > cheers, > -MH > > -- > Michael Harms, Ph.D. > ----------------------------------------------------------- > Conte Center for the Neuroscience of Mental Disorders > Washington University School of Medicine > Department of Psychiatry, Box 8134 > 660 South Euclid Ave. Tel: 314-747-6173 > St. Louis, MO 63110 Email: [email protected] > > From: <[email protected]> on behalf of Vanessa K > <[email protected]> > Date: Thursday, April 27, 2017 at 4:30 AM > To: "[email protected]" <[email protected]> > Subject: [HCP-Users] Fwd: Using multiple cores while running Structural > Pipelines > > Alright, thanks! And just for clarification for dummies: if not installing > something like grid_engine, will running all 50 subjects in one Terminal > window or 10 subjects in each of five different windows be more efficient? Or > doesn't it make any difference? And I estimate around 20 days for 50 subjects > (for the structural pipelines), is it a normal processing time? > > Thanks again! > > > 2017-04-26 16:20 GMT+02:00 Glasser, Matthew <[email protected]>: >> I find it helps to install something like gridengine which interfaces with >> fsl_sub. You can then queue up the jobs and they will all run >> automatically. You can choose how many to run at a time. I believe you >> don’t need to specify an openmp flag unless you want fewer than all of the >> cores. >> >> Peace, >> >> Matt. >> >> From: Vanessa K <[email protected]> >> Date: Wednesday, April 26, 2017 at 4:39 AM >> To: Matt Glasser <[email protected]> >> Cc: "[email protected]" <[email protected]> >> Subject: Re: [HCP-Users] Using multiple cores while running Structural >> Pipelines >> >> Thanks for your quick answer! >> Just to make sure: it means I shouldn't specify -openmp 5 in the recon-all >> line and that the multi-threading happens automatically? >> So given that I have around 50 subjects who I need to run this on, would the >> best way to accelerate the performance be to just open several terminal >> windows (e.g. 5) and run ten subjects in each? Or would it make no >> difference from running all 50 in one window? I considered looking into GNU >> parallel, but not sure it would make sense. What would your advice for such >> a number of subjects be? >> >> Thanks so much for your help. >> >> V >> >> 2017-04-25 16:55 GMT+02:00 Glasser, Matthew <[email protected]>: >>> PreFreeSurfer uses mostly single threaded tools. Some of FreeSurfer is >>> multi-threaded (and importantly one binary has some issues with >>> oversubscribed cores leading to much longer execution times). >>> PostFreeSurfer is largely multi-threaded. In all cases, the pipelines use >>> multiple threads by default if they can. >>> >>> Peace, >>> >>> Matt. >>> >>> From: <[email protected]> on behalf of Vanessa K >>> <[email protected]> >>> Date: Tuesday, April 25, 2017 at 7:36 AM >>> To: "[email protected]" <[email protected]> >>> Subject: [HCP-Users] Using multiple cores while running Structural Pipelines >>> >>> Dear users, >>> >>> it might be a silly question but I hope you can help. I have a computer >>> with 8 cores and I would like to use 7 of them while running pipelines that >>> take a lot of time (such as FreeSurfer). Is there any way to specify the >>> core number when running the script? Or are there any tricks to speed up >>> the processing time? >>> >>> Thanks a lot! >>> >>> V >>> >>> _______________________________________________ >>> HCP-Users mailing list >>> [email protected] >>> http://lists.humanconnectome.org/mailman/listinfo/hcp-users >>> >>> >>> The materials in this message are private and may contain Protected >>> Healthcare Information or other information of a sensitive nature. If you >>> are not the intended recipient, be advised that any unauthorized use, >>> disclosure, copying or the taking of any action in reliance on the contents >>> of this information is strictly prohibited. If you have received this email >>> in error, please immediately notify the sender via telephone or return mail. >> >> >> >> The materials in this message are private and may contain Protected >> Healthcare Information or other information of a sensitive nature. If you >> are not the intended recipient, be advised that any unauthorized use, >> disclosure, copying or the taking of any action in reliance on the contents >> of this information is strictly prohibited. If you have received this email >> in error, please immediately notify the sender via telephone or return mail. > > > _______________________________________________ > HCP-Users mailing list > [email protected] > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you are > not the intended recipient, be advised that any unauthorized use, disclosure, > copying or the taking of any action in reliance on the contents of this > information is strictly prohibited. If you have received this email in error, > please immediately notify the sender via telephone or return mail. > _______________________________________________ > 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
