Parallel computing out of memory

I just started my trial of Parallel Computing Toolbox, unfortunately, with little success. Now I can see that my program is utilizing the full extent of my CPU (duo Core i7) and RAM (16 GB), but it keeps crashing on me for errors of "out of memory".

Why can't Parallel Computing automatically detect the capacity of my RAM and run it accordingly? How can I specify that parloop so that it will run at the limit of my computer but without crashing?

This is my first time to use Parallel Computing. Many thanks in advance for your help.

[Leon]

Thanks for the information!

Answer 1

[OCDER]

It's possible you have a lot of broadcast variables, which are sent to individual workers for processing despite being unused, and thus they take up memory. There could also be another issue with the code itself generating a large matrix. Does it work okay without parfor? Do you know where this memory error occurs in your code? If so, provide that code for us so we can help.

Data.A = rand(4000); %This becomes a broadcast variable
parfor j = 1:4000
  Sum(j) = sum(Data.A(:, j)); %Data is sent to ALL workers, 2 extra copies  
end

Data = rand(4000) %This becomes a sliced variable. Okay!
parfor j = 1:4000
  Sum(j) = sum(Data(:, j));   %Okay! Only minimum data is sent to workers.
end

Read these:

https://www.mathworks.com/help/distcomp/troubleshoot-variables-in-parfor-loops.html

https://www.mathworks.com/help/matlab/matlab_prog/resolving-out-of-memory-errors.html