CompVision

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.

<p> MATLAB creates a local process pool with some default name. If you fire up a second copy of MATLAB on the same machine, it will use that same default pool so the two different copies will compete/conflict. </p> <p> Here is a bit of MATLAB code that will create a pool whose info is stored in a specified directory and hence won't conflict: </p> <pre class="code">mkdir('/tmp/fowlkes.1'); schd = findResource('scheduler', 'configuration', defaultParallelConfig); schd.DataLocation = '/tmp/fowlkes.1' matlabpool(schd); </pre><p> </p> <p> It should be noted that when a process pool is started, each worker switches to single-threaded mode. Meaning workers will not be able to take advantage of multithreaded computations built into many of the MATLAB functions. </p> <p> One way to get around this, is to run the command <code>pctRunOnAll maxNumCompThreads(#)</code>. This command will be executed on all workers in the pool and specifies the max number of threads each worker is allowed to use. Using the function <code>maxNumCompThreads</code> is really just a stop gap measure as it is slated for removal in the future. A better solution might be to just create a process pool right before using parfor and then tearing it down right after finishing. </p>