How to check for number of vacant workers available in a parallel pool?

I have a branch and bound algorithm that needs to be executed in parallel. The main 'if' condition that I need to implement requires me to have the number of vacant workers available in the parallel pool. Numlabs returns total number of workers in the pool. But how to obtain the number of vacant workers available in the pool in both parfor and spmd blocks?

[Edric Ellis]

It's not clear to me quite what you're asking for here. At the MATLAB client, you can find out the total number of workers by getting the NumWorkers property of the parallel pool object returned by gcp. Both parfor and spmd execute synchronously - so from the client's perspective, either all workers are busy, or none are.

If you use instead parfeval, then the client is not blocked by the execution on the workers, and there you might want to find out how many workers are currently free. This can be inferred from the properties of the FevalQueue property of the parallel pool.

[Viswanath Hariharan]

Let's say I have 100 workers to use. I send problems in batches of 4 to be solved. The condition will be that if there are at least 4 workers available, I'll send the next batch of 4. Else I'll wait. What do you think I should be using to accomplish this?

Answer 1

[tommsch]

I wrote a quick and dirty function, which returns whether there are workers available for a parfor loop. It can not return the number of workers available, nor do I know how well it works. You can find it on the File Exchange:

https://de.mathworks.com/matlabcentral/fileexchange/168896-can_execute_parfor