I use parpool and parsim, but only 1 worker 'works' instead of 4.

I have this Simulink file that runs for 3.84 s. Each 0.04 s. the data changes. I want to use parsim to make the simulation faster. I wrote the code below, but it seems that it uses only one worker (at least only one simulation appears in the manager). Is there any change I should do, or this is the correct way, because its slower than the simple sim command.

% Create a parallel pool if not already available
if isempty(gcp('nocreate'))
    parpool; % Create a parallel pool
end

% Load Model & Simulation Input
model = 'zigzag_example_matlab';
Simulink.BlockDiagram.buildRapidAcceleratorTarget(model);
simIn = Simulink.SimulationInput(model);

%Simulation

tic
simout = parsim(simIn, 'TransferBaseWorkspaceVariables', 'on', 'ShowSimulationManager', 'on');
toc

[Edric Ellis]

parsim allows you to run multiple simulations in parallel - you should split your data up into pieces, and then run each piece as a separate simulation.

[Dimosthenis Michaloudis]

You mean like 4 different models? Each one running 0.96? Wouldn't that need also 4 parsims, which would make the thing slower?

Or is it something else entirely?

Answer 1

[Walter Roberson]

You appear to be building a scalar simIn

You should be constructing an array of simulation input objects

The parsim documentation has the example,

FeedTempSweep = 250:10:300;
for i = length(FeedTempSweep):-1:1
in(i) = Simulink.SimulationInput('CSTR');
in(i) = in(i).setVariable('FeedTemp0',FeedTempSweep(i));
end
out = parsim(in, 'ShowSimulationManager', 'on')

Here, 'CSTR' is the name of the model that is to be run.

Each member of the in array defines one simulation to be run. parsim() runs several of the entries in parallel.

[Dimosthenis Michaloudis]

Hi Walter Roberson,

your help was invaluable, and it did the trick. For some reason I skipped the section in this documentation.

Although I have reworked the code, now it is far slower. Slower than the original code, and slower than by just using a sim command.

% Create a parallel pool if not already available
if isempty(gcp('nocreate'))
    parpool; % Create a parallel pool
end

% Load Model & Simulation Input
model = 'zigzag_example_matlab';
Simulink.BlockDiagram.buildRapidAcceleratorTarget(model);


%step time
Sweep = 0.00:0.04:3.84;
for i = length(Sweep):-1:1
simIn(i) = Simulink.SimulationInput(model);
simIn(i) = simIn(i).setVariable('SampleTime',Sweep(i));
end

%Simulation

tic
simout = parsim(simIn, 'TransferBaseWorkspaceVariables', 'on', 'ShowSimulationManager', 'on');
toc

[Walter Roberson]

When you use parsim(), each sweep is going to use another member of the pool. Each member of the pool is likely to be using only one core. Your model probably benefits from having more than one core available.

[Dimosthenis Michaloudis]

My PC has 4 cores, and parpool indeed creates 4 workers. But still the running time is very slow. To be more specific:

Sim command: 90 seconds

Parsim command: 5000 seconds

I guess that something is off in my code, but I can't find or understand what.

[Edric Ellis]

It does indeed sound like something isn't quite right there. Are you comparing the parsim time to something like this?

Sweep = 0.00:0.04:3.84;
for i = length(Sweep):-1:1
    SampleTime = Sweep(i);
    [t, y] = sim(model);
end

[Dimosthenis Michaloudis]

I compare the parsim to something far simpler. This:

model = 'zigzag_example_matlab.slx';

tic
simout = sim(model,'SimulationMode','rapid');
toc