parpool fails in deployed application

I am building a parallel for loop (parfor) based deployed application. I have a parpool command before the parfor loop:

parpool('local', 4);

parfor a = 1:N ....

The code works with in MATLAB interacvtive, but on the deployed application, it errors out at the parpool line:

Error using parpool (line 113) The specified superclass 'parallel.internal.customattr.CustomPropTypes' contains a parse error, cannot be found on MATLAB's search path, or is shadowed by another file with the same name.

[Nikilesh Chilkuru]

Can you tell me: 1) Which version of Matlab are you using? 2) Are you trying to run the deployed application on the same system?

[Vinayak Phalke]

MATLAB version is 2018a. The build computer and deployed computer (where the app is run) are two differing ones.

[Kevin Chng]

What is your deployed application?MATLAB executable file?

I think you might need to set your parpool profile

https://www.mathworks.com/help/compiler_sdk/cxx/deploy-applications-created-using-parallel-computing-toolbox.html

However, in documentation, it is for MATLAB Compiler SDK.

Hi Nikilesh Chilkuru,

I would like to know that does MATLAB compiler need to do the same way?

[Vinayak Phalke]

Based on your suggestion, I saved my local profile (local.settings) and bundled that file in the deployed app. That seems to have fixed this issue. Thanks

[Kevin Chng]

Great. That's mean for MATLAB compiler, it need to do the same way as MATLAB Compiler SDK.

Answer 1

[Kevin Chng]

To use parallel computing for standalone application (MATLAB Compiler SDK/MATLAB Compiler) , you may need to configure/change your code as below:

https://www.mathworks.com/help/compiler_sdk/cxx/deploy-applications-created-using-parallel-computing-toolbox.html

Answer 2

[Kristoffer Walker]

Folks,

It would be great if this answer could be improved. I am having this issue now I believe, but I do not understand what a "local profile" is, how to generate such a settings file, and why I should bundle it for deployment in a completely different remote machine that will have a different number of cores and may not share any similarities with the local profile that was deployed with it. I have read the website links above.

Kris

[Kevin Chng]

Hi Kris,

No worries, I think it is working with different PC up to 512 cores, provided it is utilizing the local computing resources. Not cloud/server resources (MATLAB Parallel Server).

How to get the setting files from local profile?

On top of your tab (Home>Parallel>Export)

Then you will get the setting file.

[Kristoffer Walker]

Hi Kevin,

By the way, is there a way to have Matlab email me when someone has commented on a post I create? I don't seem to have that option turned on and I'd like to.

I am not using the Matlab Parallel Server. I just have a GUI that has a number of threads parameter. The callback on the nThreads parameter executes a parpool. The problem is that for some choices of nThreads (still below the max number of threads available to Matlab), the deployed version of the GUI will "beep" during the callback. The non-deployed version will not beep. No console errors are observed however. I interpret the beep to be due to parpool encoutering an issue. However, it still registers as reserving the correct number of threads. The callback is extremely simple:

function nThreads_Callback(hObject, eventdata, handles)
% hObject    handle to nThreads (see GCBO)
% eventdata  reserved - to be defined in a future version of MATLAB
% handles    structure with handles and user data (see GUIDATA)

set(handles.Wavefield,'enable','off');
numCores = str2double(get(hObject, 'string'));
delete(gcp('nocreate'));

if (numCores > 1)
    f = waitbar(0,'Creating parallel processing thread pool...');
    parpool('local', numCores);
    waitbar(0.5, f, 'Creating parallel processing thread pool...');
    waitbar(1, f, 'Creating parallel processing thread pool...');
    close(f);
end

set(handles.Wavefield,'enable','on');

[Kevin Chng]

Hi Kris,

Let me provide some documentation here for your reading:

https://www.mathworks.com/help/compiler/use-the-parallel-computing-toolbox.html

Similar question has been asked in the past:

https://www.mathworks.com/matlabcentral/answers/92641-is-it-possible-to-use-parfor-in-a-compiled-application-on-my-parallel-configuration

I tried it before personally, it is working for deployed application. For your question about :

Different remote machine that will have a different number of cores and may not share any similarities with the local profile that was deployed with it.

Frankly speaking, i'm not sure about it, but i think the local setting file should working for different machine.

follow the documentation to get your local pc work with your deployed standalone application with parallel computing first.

https://www.mathworks.com/help/compiler/use-the-parallel-computing-toolbox.html

later, we can try to execute the standalone application in different pc. I think it should be working. Let me know, i'm interested to know it as well.

If you still encouter any issue, let me know, then i try spend some time to try it out.

Answer 3

[Kristoffer Walker]

Hi Kevin,

So I've looked more carefully at this and created a simple example to demo the problem. Just create a script "testApp.m". Include in it a call to function testAppFunc(), which inside it only does a "parpool(2)". Put some fprintf statements in there for correlation purposes. Then compile it with Application Compiler and test the testApp.exe in for_redistribution_files_only directory by launching from a PowerShell. You will hear the beep. Why does this happen and how can I stop it? This only happens the first time parpool is executed from a deployed app. It does not happen during subsequent calls to parpool in the same instance of the deployed app. It also does not happen in the non-deployed app at any time.

BTW, saving the settings.local file did not help (but thanks for those links to the instructions).

Kris