Is it okay to use matlabpool in multiple matlab sessions?

I'm sharing a server with someone else in my lab. To reduce executive time, I use the matlab parallel toolbox. However, matlab gives me this warning:

Warning: Found 1 pre-existing parallel job(s) created by matlabpool that are running.

This is because another person is also using the matlabpool. Should I ignore this warning? Or should I better use matlabpool when others don't? Need some help here!

[Matt J]

Are you sure that the job mentioned in the warning is due to the other sesssion? What happens when you execute "matlabpool close" and then try to re-open the pool?

[Peng]

I'm pretty sure. I just did what you said and the warning came up again.

[Matt J]

Can you ask that person to close his/her matlabpool to test whether the problem goes away?

[Peng]

I tried that too. Everything is fine as long as there is just one matlab session which is using the matlabpool.

Answer 1

[Matt J]

OK. Well, since you're sharing parallel computing resources with others, you'll obviously get better performance if you're the only one using the pool. I can't imagine it will affect the validity of your computations, though, if that's what you're worried about, and the fact that you're getting a warning vs. an error seems to reinforce that.

You can always try and see what happens.

Answer 2

[Peng]

Problem solved! Higher version of matlab doesn't have this warning. Maybe its time to use r2012b.

Answer 3

[Thomas Ibbotson]

By default matlabpool will try and use all the workers available in a cluster. For example if you have 16 workers in a cluster and open a matlabpool:

>> matlabpool open myClusterProfile
Starting matlabpool ... connected to 16 workers.

Now the entire cluster is busy with your interactive matlabpool session. Someone in another MATLAB session will not be able to use the cluster (including opening their own matlabpool) until you close yours. It is not possible to share a single matlabpool between matlab sessions.

You can however share the cluster, by not using all the available workers, e.g.

>> matlabpool open myClusterProfile 8
Starting matlabpool ... connected to 8 workers.

Now there are still 8 workers available for someone else to use. Either with their own matlabpool or with batch jobs.

HTH, Tom