How to send a variable to the worker workspace?

An example code is show below:

classdef myclass
    properties
        
    end
    
    methods
        function obj = myclass
            obj.variable1 = dsm.signal.Bias(3);
            theClass = class(obj)
            assignin('base','theClass',theClass)
        end
    end

% When loading the class and run by using parsim(), the system can't find
% theClass in the worker's workspace. 

% My question is: how to send "theClass" into the parallel worker's
% workspace? 

[Rui Zhang]

When the parallel runs with one worker, the system can find "theClass" in the worker's workspace and there is no error in the simulation.

When the paralle runs with 2 workers, the system can't find "theClass" in the worker;s workspace when simulation starts. An error came.

Answer 1

[Jacob Mathew]

Hey Rui,

To send the “myclass” class that you defined, to the worker’s workspace, you can use the “addAttachedFiles” function. The documentation for the function can be found at the following link:

https://www.mathworks.com/help/releases/R2023b/parallel-computing/parallel.pool.addattachedfiles.html

Here is an example of using the function with a pool of 3 workers. The “myclass” is the same as you have provided, but with “variable1” added as property of the class. We then create the pool, then add the class file to it and then every worker creates an object of the class:

classdef myclass
    properties
        variable1
    end
    
    methods    
        function obj = myclass
            obj.variable1 = 3; % Your property value here
            theClass = class(obj);
            assignin('base','theClass',theClass)
        end 
    end 
end

% Assuming the class definition is in a file named myclass.m
% Clear and stop any previous pools. Then create a pool of 3 workers

% clear all
% delete(gcp('nocreate'))
pool = parpool(3);

% attach the file to be workers
addAttachedFiles(pool,{'myclass.m'})

% Use the spmd block to execute code on each worker
spmd  
    % Create an instance of myclass on each worker
    obj = myclass;
    
    % Display the variable1 value in the object
    disp(['Worker ', num2str(spmdIndex), ' myclass objects value of variable1: ', num2str(obj.variable1)]);
end

We get the output as follows:

>>
% Starting parallel pool (parpool) using the 'Processes' profile 
%     connected to parallel pool with 3 workers.
% 
% Worker 1: 
% Worker 1 myclass objects value of variable1: 3
% 
% Worker 2: 
% Worker 2 myclass objects value of variable1: 3
% 
% Worker 3: 
% Worker 3 myclass objects value of variable1: 3

However, if you don’t want to send the class file to the workers and instead want to copy the object between the worker’s worspace, then we can use “parallel.pool.Constant” method. The documentation for the method can be found in the link below:

https://www.mathworks.com/help/releases/R2023b/matlab/ref/parallel.pool.constant.html

Using this, you can share copies of the object itself to the worker’s workspace. The following is a code snippet that does this:

% Assuming the class definition is in a file named myclass.m
% Clear and stop any previous pools. Then create a pool of 3 workers

% clear
% delete(gcp('nocreate'))
pool = parpool(3);

% creating an object of the class
theClass = myclass;

% storing the object
distributedObj = parallel.pool.Constant(theClass);

% Use the spmd block to execute code on each worker
spmd
    % getting the object
    obj = distributedObj.Value;
    
    % Displaying the value of variable1 in object
    disp(['Worker ', num2str(spmdIndex), ' accessing myClass variable1: ', num2str(obj.variable1)]); 
end

We get the output as follow:

>> 
% Starting parallel pool (parpool) using the 'Processes' profile
%     Connected to parallel pool with 3 workers.
% 
% Worker 1: 
% Worker 1 accessing myClass variable1: 3
% 
% Worker 2: 
% Worker 2 accessing myClass variable1: 3
% 
% Worker 3: 
% Worker 3 accessing myClass variable1: 3