Integrating Parallel Computing with App Designer for Real-Time UI Updates

I’ve developed an application with three panels, each responsible for a different task, and I need to update the UI in real-time. However, I’m running into issues because App Designer doesn’t support threading directly and struggles with accessing app property variables. I would like to utilize the Parallel Computing Toolbox to enable parallel execution of tasks and update the UI simultaneously.

I’ve already tried using parfor, parfeval, and spmd to handle parallel execution, but I’m facing difficulties with updating the app property variables in real-time within the App Designer environment. Specifically, these approaches are not integrating well with the app's UI and property variables.

Can you provide guidance on how to properly integrate parallel computing with App Designer while ensuring the app property variables are updated correctly in real-time? Any tips or solutions to resolve this issue would be very helpful!

Answer 1

[Rick Amos]

My recommendation is to use a mix of parfeval (to launch parallel work) with afterEach/afterAll (to update the app in response to that work finishing):

• https://uk.mathworks.com/help/matlab/ref/parallel.future.aftereach.html
• https://uk.mathworks.com/help/matlab/ref/parallel.future.afterall.html

You are allowed to update the App inside the afterEach/afterAll function. This means you can think of these as like Callbacks to parallel work.

For an example, create an AppDesigner app named "MyApp" with a button and a UIAxis. Then add the following code:

classdef MyApp < .. % AppDesigner app with a "Start" button and a single UI Axes

    methods (Static)
        function result = doSomeWork()
            pause(1);
            result = rand(100, 1);
        end
    end
    
    methods
        % One of the doSomeWork tasks finished with a result
        function OneTaskFinished(app, result)
            hold(app.UIAxes, "on");
            plot(app.UIAxes, result);
            hold(app.UIAxes, "off");
            drawnow("limitrate", "nocallback");
        end
        
        % All tasks finished. This receives the futures because
        % PassFuture=true in the above afterAll
        function AllTasksFinished(app, futures)
            % Report the first error if one is present
            for ii = 1:numel(futures)
                if ~isempty(futures(ii).Error)
                    errordlg(getReport(future(ii).Error));
                end
            end
            %... Make the App look like it's idle again
            app.Button.Enable = true;
        end
    end
    
    methods
        % Callback for a Start button (add callback through AppDesigner)
        function ButtonPushed(app, event)
            %... Make the App look like it's running
            app.Button.Enable = false;
            % Launch some parallel work
            for ii = 1:10
                futures(ii) = parfeval(@MyApp.doSomeWork, 1);
            end
            % Setup an app callback run per task
            afterEach(futures, @app.OneTaskFinished, 0);
            % Setup an app callback run once all work finishes
            % PassFuture allows us to understand if any task errored.
            afterAll(futures, @app.AllTasksFinished, 0, PassFuture=true);
        end
    end
end

[BHARATH]

Hi @Rick Amos,

Can we able to call an argument for function doSomeWork() and return that in a EditField.

[Walter Roberson]

@BHARATH

Currently the call is

futures(ii) = parfeval(@MyApp.doSomeWork, 1);

You can add additional parameters after the 1 such as

futures(ii) = parfeval(@MyApp.doSomeWork, 1, MyApp.SomeEditField.Value);

You would need to alter the function signature

function result = doSomeWork()

to expect the parameter.

You can alter function OneTaskFinished to set an edit field. However, function OneTaskFinished will be invoked several times (once for each "future" created) and it is questionable which edit field you should set. I guess you could just append each result to the end of an edit field.

[BHARATH]

Hi @Walter Roberson,

Thanks for your answer earlier, but I got struck at MATLAB App Designer with four tabs, each handling a distinct process:

Tab 1: Processes data from a Analog Input and Output interface from DAQ.

Tab 2: Manages communication over a CAN protocol.

Tab 3: Handles data imports and processing data from excel sheet.

Tab 4: Provides a UI for controlling and monitoring the other tabs.

In Tab 4, a button triggers the processes in Tabs 1, 2, and 3 to start one after the other. However, only the first two processes are triggering successfully, while the third one doesn’t start. Each process runs continuously, and I suspect the issue may be due to blocking or conflicts between the running processes.

Questions:

1. How can I ensure all processes trigger sequentially without one blocking the other?
2. What is the best way to handle data synchronization across continuously running processes to prevent conflicts?

Any guidance or examples for managing synchronization across multiple tabs would be great.

Thank you!

[BHARATH]

Hi @Walter Roberson, Is there anyway we can do this?

[Walter Roberson]

However, only the first two processes are triggering successfully, while the third one doesn’t start.

That could happen if your parpool size comes out as 2.

[BHARATH]

@Walter Roberson,Sorry I'm not using parallel pool at all as of now. I’m running 4 processes, including multiple APIs that I can’t directly parallelize because they involve DAQ and CAN operations. Is there a way to implement these in a parallel process?

[Walter Roberson]

You wrote, "Each process runs continuously". That implies you are using parallel processing. Parallel processing requires either the use of Parallel Computing Toolbox, or else the use of Background Pools. Your code uses parfeval but does not construct a background pool, so you must be using the Parallel Computing Toolbox.

parfeval() automatically opens a parpool if there is no existing parallel pool.

[BHARATH]

@Walter Roberson I should have clarified this earlier. For the CAN functions, they are called by a timer function, which triggers every 1 millisecond. For the DAQ functions, they are invoked through a callback in the Data Acquisition Toolbox, which is triggered after receiving a set of data.

Additionally, I have experimented with Parallel Computing Toolbox functions like parfeval(), parfevalOnAll(), and spmd in my code. My goal was to call button-pressed callback functions from App Designer using these parallel processing functions.

I’m sharing the code I used to create the parallel pool for reference. Let me know if you have suggestions

% Start a parallel pool if none exists
if isempty(gcp('nocreate'))
    parpool(); % Start a parallel pool
end

Starting parallel pool (parpool) using the 'Processes' profile ...

Error using parpool (line 133)
No default cluster available. Create a cluster and its associated profile.

% Get the current parallel pool
p = gcp();

% Run tasks on different workers
f1 = parfeval(p, @canRead, 0, app, 1); 
f2 = parfeval(p, @ReadIoFromExcel, 0, app);
f3 = parfeval(p, @StartInputAcquisition, 0, app);
f4 = parfeval(p, @StartOutputAcquisition, 0, app);

% Optionally fetch results asynchronously
results = cell(4, 1);
results{1} = fetchOutputs(f1);
results{2} = fetchOutputs(f2);
results{3} = fetchOutputs(f3);
results{4} = fetchOutputs(f4);

Answer 2

[Walter Roberson]

The only Mathworks product that is designed for Real-Time work is Simulink Real-Time.

You will never be able to get Real-Time performance from MATLAB (including App Designer.) MATLAB is not designed for Real-Time and will never be designed for Real-Time.