how to run several tasks in a job in parallel ?

I have a problem about functions createJob and createTask. The problem is, I created 3 tasks in a job, I expected that they can be run in parallel. But actually they can't.

I have a function jobTest2.m, which returns labindex of worker.

code of jobTest2.m:

function [a] = jobTest2
% fprintf('labindex: %d\n',labindex);
a = labindex;
pause(20);
end

code run in matlab command window:

parallel.defaultClusterProfile('local');
c = parcluster();
j = createJob(c)
createTask(j, @jobTest2, 1, {});
createTask(j, @jobTest2, 1, {});
createTask(j, @jobTest2, 1, {});
submit(j);
wait(j);
results = fetchOutputs(j)

output:

results = 

      [1]
      [1]
      [1]

It takes 3 times longer than just running a task. And the labindex shows that it runs in serial. I've read a matlab answers Parallel programming with createJob and followed its proposal. But the problem is still not solved.

Before I run these code, I've delete all jobs on the cluster.

>> delete(c.Jobs)
>> c

    c = 

   Local Cluster

      Properties: 

                            Profile: local
                           Modified: false
                               Host: localhost
                         NumWorkers: 4

                 JobStorageLocation: /home/drinkcor/.matlab/local_cluster_jobs/R2015b
   RequiresMathWorksHostedLicensing: false

      Associated Jobs: 

                     Number Pending: 0
                      Number Queued: 0
                     Number Running: 0
                    Number Finished: 0

Can anyone help me? Thanks in advance!

[Dingqiao Zhu]

Besides, I didn't open any parpool here.

Answer 1

[Dingqiao Zhu]

I've found the solution. As @Edric Ellis reply in matlab ask Why is createJob / createTask so much slower than parfor? ,

"The thing to remember about your timings is that when using createJob and createTask with the local cluster type, each task runs in a separate MATLAB process. This takes much more time to launch than a parfor loop.

I suggest looking into parfeval, which has a similar interface to createTask, but uses the workers in a parallel pool, and so is much more efficient."

So, the solution is,

function parfevalTest

p = gcp();

tstart = tic;

f1 = parfeval(p,@jobTest2,1);
f2 = parfeval(p,@jobTest2,1);

value1 = fetchOutputs(f1);
value2 = fetchOutputs(f2);

aaaatime = toc(tstart);
fprintf('Time Consuming: %f\n',aaaatime);
fprintf('labIndex of job1: %f\n',value1);
fprintf('labIndex of job2: %f\n',value2);

end

The result is,

>> parfevalTest
Time Consuming: 20.063211
labIndex of job1: 1.000000
labIndex of job2: 1.000000

Although their labindex are the same, they do run in parallel according to the time consuming. The value of labindex is always 1 both in parfor and parfeval.