Running >1 matlabpool jobs with the same local pool

2 views (last 30 days)
Eng Hee Yeo
Eng Hee Yeo on 11 Jul 2014
Commented: Eng Hee Yeo on 14 Jul 2014
I have a user who is running parallel jobs using the 'local' matlabpool profile and is encountering problems when running 2 jobs concurrently. He knows that the local system can have 12 workers in the local pool, and so he divided the 2 jobs using 6 workers each, but with the same command:
matlabpool close force local
CoreNum=6;
matlabpool('open','local',CoreNum);
This seems wrong, but I cannot find anywhere in the documentation to verify my suspicions. Can anyone advise?

Answers (1)

Shashank Prasanna
Shashank Prasanna on 11 Jul 2014
Maximum of 1 MATLAB worker per CPU core is recommended. This does not include hyperthreaded virtual cores. As an examples, if you have a dual core i5 processor, then it is recommended you start a local pool of 2 workers. Any more workers will mean they fight for cpu time and you most likely won't see any gain in performance.
  1 Comment
Eng Hee Yeo
Eng Hee Yeo on 14 Jul 2014
Thanks Shashank, but this is not what I am asking. My server has 12 cores, so, 2 x 6 workers is fine and there is no resource contention. My original question is whether it is correct to run 2 separate jobs using the same local matlabpool, with 6 workers each. What happens if one of the jobs issues a command like "matlabpool force close local"? Will the other job be affected?

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!