Problem working with parfor

Dear all,

I am working with the parallel tool in order to optimize the run time of my simulations. I am using a resevoir simulator (MATLAB Reservoir Simulation Toolbox). I am adding a function after each time convergence, to do a parallel calculations on each cell. However, it seems there is a problem with the parallel tool. It stuck in a cetrain function as found in the attached image.

The code stucks here does not proceed anymore. Do you have any idea why it is happening?

Answer 1

[Edric Ellis]

That stack frame in the profiler is simply where your MATLAB client is waiting for the workers to complete their work. You can get profiling information from the workers using mpiprofile.

[Ahmadreza]

I found that the code is stucked in below function:

        function [tags, results] = getCompleteIntervals(obj, numIntervals)
            
            tags = nan(numIntervals, 1);
            results = cell(numIntervals, 2);
            for i = 1:numIntervals
                err = [];
                r = parallel.internal.pool.ParforIntervalResult();
                while r.isEmpty()
                    assert(obj.NumIntervalsInController > 0, ...
                        'Internal error in PARFOR - no intervals to retrieve.');
                    r = obj.ParforController.waitForNextCompletedInterval(obj.AwaitIntervalPollPeriod);
                    
                    % In each poll tick of the polling wait, yield to any
                    % events allowed to interrupt the parallel language.
                    parallel.internal.pool.yield();
                    if r.isEmpty()
                        % Only test to see if the session is failing if we didn't get a
                        % results from the queue
                        if ~obj.isSessionValidAndRunning()
                            errorMessageInput = iGetParpoolLinkForError();
                            error(message('parallel:lang:parfor:SessionShutDown', errorMessageInput));
                        end
                    else
                        obj.NumIntervalsInController = obj.NumIntervalsInController - 1;
                        if r.hasError()
                            % Maybe try again
                            [r, err] = obj.handleIntervalErrorResult(r);
                        end
                    end
                end
                % Check to see if the interval result has an error
                if r.hasError()
                    throw(err);
                else
                    tags(i) = r.getTag();
                    data = parallel.internal.pool.optionallyDeserialize(r.getResult());
                    assert(numel(data) == 2, ...
                        'Unexpectedly received the incorrect number of outputs.');
                    results(i,:) = data;
                    obj.IncompleteIntervalsCell{tags(i)} = [];
                end
            end
            % Yield one final time to ensure all events directly generated by
            % code in the PARFOR loop are executed before we give control back
            % to the user.
            parallel.internal.pool.yield();
        end

It is stucked in the while loop, and it does not go further as r is empty (r.isEmpty()).

[Edric Ellis]

That means the workers are still working on your loop. The simplest way to check that things are still proceeding on the workers is to add a disp statement to the body of your parfor loop to see where things are getting to.

[Ahmadreza]

Thanks for your help. As a matter of fact you are right. There was a problem inside the third party software that I fixed it. I MUST NEVER LOSE MY FAITH IN MATLAB!