How to run a simulink model in rapid accelerator on a cluster?

I am trying to calculate a simulink model in parallel on a cluster. According to serveral sources [1,2] from the internet the simulink model should be executed in rapid accelerator to run faster. However, all examples which I have found are running with the profile parpool('local').

Update: two files attached for simplicity. Please try them out.

My problem is, that my calculations works on parpool('local') but not if I am using a cluster profile. Given the following code

First I open a cluster by

    poolhdl = parpool(cluster_name, 4);

Define my tunable parameters (I have a simulink model with a sine where I change offset and amplitude. According to [1] I enable 'Inline Parameters'.)

    mdl = 'parallel_computing_sim';
    a = Simulink.Parameter;
    a.CoderInfo.StorageClass = 'SimulinkGlobal';
    b = Simulink.Parameter;
    b.CoderInfo.StorageClass = 'SimulinkGlobal';

    a.Value = 10;
    b.Value = 10;

Build my system in rapid accelerator

    %%build system in rapid accelerator
    load_system(mdl);
    set_param(mdl, 'Solver', 'ode4')
    set_param(mdl, 'FixedStep', num2str('1e-5'))
    rtp          = Simulink.BlockDiagram.buildRapidAcceleratorTarget(mdl);
    close_system(mdl, 0);

Generate a parameterset

    %%generate parameterset
    iterations = 0;
    for j = 1:3
      for i = 1:3
        iterations = iterations +1;
        paramSet{iterations} = Simulink.BlockDiagram.modifyTunableParameters(rtp, 'a',i, 'b', j);
      end
    end

and perform the parfor loop

    load_system(mdl)
    parfor i = 1:iterations
      simout{i} = sim('parallel_computing_sim',...
      'SimulationMode','rapid',...
      'RapidAcceleratorUpToDateCheck','off', ...
      'RapidAcceleratorParameterSets',paramSet{i});
    end;

Then I get the error message:

    Unable to load block diagram 'parallel_computing_sim' (line xyz)

in the line of the parfor loop. I assume that the binary of the rapid accelerator is not transfered to the workers correctly.

That is why I tried something like

    addpath('./slprj/raccel/parallel_computing_sim')
    poolobj = gcp;
    addAttachedFiles(poolobj,{'parallel_computing_sim'})

Then I get the error message:

    Error using parallel_computing_sim_start (line xyz)
    Executable not found.

I valided the cluster by 'Validate' under 'Cluster Profile Manager' and it passed all tests.

Has anybody an idea how to calculate the cluster in rapid accelerator on clusters?

Thanks a lot!

[1] http://blogs.mathworks.com/simulink/2015/03/19/getting-the-most-out-of-rapid-accelerator-mode

[2] https://de.mathworks.com/help/simulink/ug/running-parallel-simulations.html

Answer 1

[Sanjay Nair]

Hello,

A big challenge with regards to parallel simulations with clusters is that the hosts in the cluster will not share the same file system as the client. This is why it's easier to solve the parallel simulations problem using a 'local' cluster profile as opposed to an 'mjs' cluster profile.

The easiest solution to your problem would be to upgrade your version of MATLAB to R2017a and make use of the parsim command: https://www.mathworks.com/help/simulink/slref/parsim.html

This command is meant as a quick access tool to running parallel simulations and supports both 'local' cluster and 'mjs' workflows.