SimEvents: Parallel Simulations produce same results for each MatLab worker

Hello,

I am trying to run the following:

function means = basic_parallel_test()
modelname = 'G_G_1_5';
reps = 10;

means = ones(10,1);

parfor j = 1 : reps
    load_system(strcat(modelname,'.slx'));
    se_randomizeseeds(modelname);
    simout = sim(modelname);
    y = simout.get('ct')
    means(j,1) = mean(y.signals.values(:,1))
    bdclose(modelname);
end;

end

No errors are reported. So far so good...

But the result is:

ans =

   28.3941
   30.0470
   29.8563
   27.9616
   30.0470
   29.8563
   27.9616
   27.9957
   28.3941
   27.9957

Every value is there twice. So it seems since I had 2 matlabworkers, both accessed the same instance of simout. Sice I defined that variable in the parfor-loop I thought it would only be visible inside the loop.

How could I do this right?

Thanks Daniel

Answer 1

[Edric Ellis]

The reason for the duplicate results is that your call to 'se_randomizeseeds' is using the system clock to set up the random number generators. So, because the workers are in sync, they're both getting the same value. You can get around this by explicitly passing a 'GlobalSeed' option to 'se_randomizeseeds'. Here's one way to do that, where we're using a combination of the current time:

parfor j = 1 : reps
  seed = mod(floor((j/reps) * now * 8640000),2^31-1);
  se_randomizeseeds(modelname, 'GlobalSeed', seed);
  ...
end

Answer 2

[Daniel Huber]

Thanks, that was it.

I totally forgot about my question and somehow I expected to get an email, when someone answers...