parallel run keps incresing RAM usage

I have a conceptually very simple matlab script to run a simulink model to explore different param combinations:

% global param settings
a=
b=
c=

% iterations on "local" params and run N instances in parallel
for d ...
    for e ...
        for f ...
            simIn(1:Nproc) = Simulink.SimulationInput(model);
            for idx = 1:N
                simIn(idx) = simIn(idx).setBlockParameter(...)
            end
            out = parsim(simIn, TransferBaseWorkspaceVariables='on')
        end
    end
end

The problem is that while iterating over d,e,f (i.e., for each parallel run) the RAM used increases until the runs start to use swap space, dramatically reducing performances.

I tried to open the simulink model (and close it) before (and after) every parallel run.

I tried to add Simulink.sdi.setRecordData(false) just after opening the model.

Now I have no idea of why the problem happens and how to solve it.

[Walter Roberson]

You appear to be overwriting the results stored in out upon each iteration of for f .

If you are saving the results somehow instead, then the details of how you are doing that might be important.

[alberto maurizi]

Well, actually I save the part of the output I need in some files (I don't keep any extra variable in memory) and then I just reuse 'out'.