Problem in running multiple parallel jobs on cluster

Hi,

I'm running a program using parallel computing toolbox on a cluster (just a simple matlabpool-parfor loop). It happened that if I only have one program running, things are fine. However, when I submit the second job (using same functions with different parameters), I got an error in one of the programs that looks like:

_??? Error using ==> matlabpool at 127

> The interactive parallel job is unavailable.

> Most likely, the parallel job has been destroyed.

I can run these multiple jobs simultaneously with no problem without PCT (just use "for").

If anyone can help explain this behavior, I would be really appreciated.

Answer 1

[Edric Ellis]

Are you including the "matlabpool" command in your code to run on the cluster? If so, that's not the right way, you should be submitting a job which already has a pool available. For example,

    numArgsOut = 1;
    argsIn = {1, 2, 3};
    job = batch( @myFcn, numArgsOut, argsIn, 'Matlabpool', 4 );

will submit a job where "myFcn" is run and can use PARFOR loops. See the batch documentation for more.

[Nada]

Hi Edric,

Thank you for your quick reply.

I followed your suggestion and the problem solved. The only thing I noticed is that when I called batch(...,'Matlabpool',3). It does use 3 workers on the cluster, not 3+1 as I expected.

Is there a further document on what's wrong with using matlabpool on the cluster?

thanks again!