Parfor Loop & EnergyPlus Co-Simulation

Hello everyone,

I'm trying to run a parfor loop to run the EnergyPlus toolbox co-simulation a bit faster (EnergyPlus Co-Simulation). I tried a lot but I'm getting an error that I'm unable to solve.

Below is my code:

ep=[];
tStart = tic;
i=0;
HVACEnergy = array2table(zeros(25,8));
Blg_Electricity = array2table(zeros(25,8));
Total_Energy = array2table(zeros(25,8));
 
parfor c=1:8
%%
t = getCurrentTask();
d = fullfile(tempdir, sprintf('worker_%d', t.ID));
if ~exist(d, 'dir'), mkdir(d); end
%%
ep = mlep;
ep.workDir = d;
% Building simulation file
ep.idfFile = 'C:\Users\mak19\Desktop\UBC_PhD\Trial_Two\6-Modeling\1-Trial_One\4-IHDX\DD_Winter_EE.idf';
% Weather file
ep.epwFile = 'C:\Users\mak19\Desktop\UBC_PhD\Trial_Two\6-Modeling\1-Trial_One\4-IHDX\CAN_BC_Kamloops.718870_CWEC.epw';
 
% Initialize
ep.initialize; 
% Simulation length
endTime = 1*24*60*60; %[s]
 
% Logging
VN = ep.outputSigName;
VN{3} = 'HVAC_Energy';
VN{4} = 'Total_Building_Electricity';
VN{5} = 'Total_Electricity';
 
logTable = table('Size',[0, 1 + ep.nOut],...
    'VariableTypes',repmat({'double'},1,1 + ep.nOut),...
    'VariableNames',[{'Time'}; VN]);
 
% Start co-simulation
ep.start;
iLog = 1;
 
%%
% The simulation loop
t = 0;
while t < endTime
    % Prepare inputs (possibly from last outputs)
    u = [20 24 1 1 1 1 1 1 1];
 
    % Get outputs from EnergyPlus
    [y, t] = ep.read;
    
    % Send inputs to EnergyPlus
    ep.write(u,t); 
    
    % Log
    logTable(iLog, :) = num2cell([t y(:)']);
    iLog = iLog + 1;
    i=i+1;
end
% Stop co-simulation process
%%
ep.stop;
%%
delete(ep)
% 
HVACEnergy(:,c) = logTable(:,4);
Blg_Electricity(:,c)=logTable(:,5);
Total_Energy(:,c)=logTable(:,6);
end
TotalTime = toc(tStart);

And this is the error I'm getting:

Error using mlep/acceptSocket (line 1144)
Java exception occurred:
java.net.SocketTimeoutException: Accept timed out
	at java.net.DualStackPlainSocketImpl.waitForNewConnection(Native Method)
	at java.net.DualStackPlainSocketImpl.socketAccept(DualStackPlainSocketImpl.java:135)
	at java.net.AbstractPlainSocketImpl.accept(AbstractPlainSocketImpl.java:409)
	at java.net.PlainSocketImpl.accept(PlainSocketImpl.java:199)
	at java.net.ServerSocket.implAccept(ServerSocket.java:545)
	at java.net.ServerSocket.accept(ServerSocket.java:513)
Error in mlep/start (line 289)
            obj.acceptSocket;

Many thanks in advance.

Mohamad

Answer 1

[Raymond Norris]

I would suggest trying this first with a pool size of 1. If it works, then my guess is that mlep only allows one connection at a time.

[Mohamad Khalil]

Thanks, Raymond. With one pool, the code is working fine and wthout errors. However, when I try to increase to number of pools gradually, it sometimes actually work (therefore I think parfor could be applied to mlep), but other time not. I am really not sure what is happening there!

[Raymond Norris]

Is there a limit to the number of connections? For instance, does it always work with 2 workers, 3 workers, ... but always fail with 5, 6, 7?

If it sometimes passes with a higher number and sometimes fails with a smaller number of workers, my guess is that all the workers are trying to ping the server at the same time and connections fail quickly. Perhaps adding a random pause will help give the delay you need. I would suggest adjusting the pause also based on the number of workers running. For example:

% Add random delay to allow connection requests from each worker
p = gcp('nocreate');
pause(rand*p.NumWorkers*5)
ep = mlep;

[Mohamad Khalil]

I implemented your suggestion, and it is working for 8 workers now, however, I can only perform 4 parallel runs, i.e. parfor c=1:4. I tried chaning the delay time, but still not working with 5+ loops! It seems as a concept the "mlep" accepts parallel computing, but there is something that prevent it from finishing the simulation. Any suggestion?

[Raymond Norris]

I'm missing something here. It works for 8 workers, but can only perform 4 parallel runs. It's not working with 5+ loops. How can it work for 8 workers, but only run 4 parallel runs? What happens on the 5 iteration? It hangs, it errors, etc.? And what do you mean that the sims are not finishing? Are you now no longer getting

java.net.SocketTimeoutException: Accept timed out

[Mohamad Khalil]

Yes, it only did up-to 4 parallel runs. When I'm doing 5+ parallel iterations, I am getting the same error again:

Error using mlep/acceptSocket (line 1144)
Java exception occurred:
java.net.SocketTimeoutException: Accept timed out
....

[Raymond Norris]

I would suggest increasing the pause for a larger number of workers

pause(rand*p.NumWorkers*15)

[Mohamad Khalil]

It actually did not work! I'm not sure what is going on. Something appears to be unstable as even with the pause code, the parallel loop crashes sometimes. I tried my code in another computer and I got a similar error.

[Mohamad Khalil]

It seems it has something to do with the "socket.cfg"! Although I'm doing as the answer in this post: https://www.mathworks.com/matlabcentral/answers/443270-parfor-energyplus-co-simulation-toolbox, I'm still getting an error. Your support means a lot for me. Thanks.

[Mohamad Khalil]

Is it possible to copy the "mlep" function to the multiple workers in a way that each worker has its own "mlep" function?. I feel might be better rather than all of the workers are using the same "mlep", which might not be supporting the "parfor" loop? Thanks.