Parallelization: yes or not?

[Daniele Marinazzo]

Hi,
We are exploring parallelization to speed things up.
We could parallelize across voxels or subjects.
Two questions: is this already implemented in other apps in an efficient way?
Then, does it make sense to parallelize within the app, or better to do it in bash when launching the app for many subjects?
Is something similar already implemented in Openneuro.org for example?
Thanks
Daniele

[Chris Gorgolewski]

Hi Daniele,

There are three levels of parallelization supported by BIDS Apps:

1) across sessions; analysis level: session with --session_label (assumes independent calls to your app which could be run on different machines/nodes)

2) across participants; analysis level: participant with --participant_label  (also assumes independent calls to your app which could be run on different machines/nodes)

3) withing single app execution (multithreading) enabled with --n_cpus flag

Most apps use 2 and 3. This is explained more at http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005209#sec013

Best,

Chris

--
You received this message because you are subscribed to the Google Groups "bids-apps-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bids-apps-de...@googlegroups.com.
To post to this group, send email to bids-a...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bids-apps-dev/2b618fac-2493-4d12-a276-cc8f1ef465a7%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Daniele Marinazzo]

Thank you Chris! Is any of these implemented/preferred for inclusion in Openneuro.org? 

[Chris Gorgolewski]

We are going through some deep refactor now but 2 and 3 were supported.

Best,
Chris

PS Apologies for brevity and potential typos. This message was composed on a phone.