How to execute parallel functions ?

9 vues (au cours des 30 derniers jours)
Paul-Henri Michel
Paul-Henri Michel le 22 Avr 2013
Hi everyones,
I have a three buffer system and It takes 3 steps : - I need to collect data from a .txt file - I need to filter these data - I need to plot the filtered data
I want to do these three steps simultaneously ! (even if I have a delay between current data and plotted data) Should I use a parfor loop, matlabpool or multi-thread programming ?
Thanks a lot !
  1 commentaire
Paul-Henri Michel
Paul-Henri Michel le 22 Avr 2013
Notice I need to have all the computed data on the same workspace, and I have to use these at each moment.

Connectez-vous pour commenter.

Réponse acceptée

Friedrich
Friedrich le 24 Avr 2013
Modifié(e) : Friedrich le 24 Avr 2013
Hi,
look at createTask and createJob. You create a Task for each file, let the worker process is and pass the data back to MATLAB. In MATLAB you get that data and plot it and restart the process:
c = parcluster; %or maybe c = findResource('scheduler','Configuration','local');
job = createJob(c);
for i=1:3
createTask(job,@rand,1,{i});
%create a function handle to the function you use to process the file, pass in the filename as argument to the function to be able to use the same function for all files
end
submit(job);
wait(job);
results = job.fetchOutputs %maybe you need getAllOutputArguments depends on the version you are using
job.delete
Or use SPMD (this should give a better peformance as CreateJob/CreateTask because te workers need to be started only once)
matlabpool open 3
result = cell(3,1);
spmd
switch labindex
case 1
result{1} = rand(1); %process file 1
case 2
result{2} = rand(2); %process file 2
case 3
result{3} = rand(3); %process file 3
end
end
%each worker holds its own result because result is a Composite object
result
result{:}

Plus de réponses (4)

Paul-Henri Michel
Paul-Henri Michel le 23 Avr 2013
No answers ?
Thanks !
  11 commentaires
Paul-Henri Michel
Paul-Henri Michel le 23 Avr 2013
Modifié(e) : Paul-Henri Michel le 23 Avr 2013
No I mean one Matlab for computing and one Matlab for plotting. Once the first Matlab has finished to filter data, I send the data to the second Matlab, which display these on a "hold on" figure and I restart the operation !
But maybe the UDP protocol is too slow to do this, because my file data contain more than 4 000 characters.
Friedrich
Friedrich le 23 Avr 2013
Plotting doesnt take much time. Using a MATLAB for Plotting only seems a bit too much. I would do all that in one MATLAB. You can use PCT to read in the 3files and post process it. You then use these results and plot it and then start over again. This can run in 1 MATLAB.
You use the second MATLAB for your own.

Connectez-vous pour commenter.


Paul-Henri Michel
Paul-Henri Michel le 23 Avr 2013
Ok, now I don't consider plotting data, I just want read data and filter data (other data obviously) simultaneously. In fact, I need two workers on the same workspace ?
Should I use parfor loop, matlabpool ?
Thanks !
  2 commentaires
Friedrich
Friedrich le 23 Avr 2013
Why do you need the same workspace? You can put the data together later on.
Paul-Henri Michel
Paul-Henri Michel le 23 Avr 2013
What do you mean by later ? On each end of cycle (end of each worker) I need data and I restart a new loop.

Connectez-vous pour commenter.


Paul-Henri Michel
Paul-Henri Michel le 23 Avr 2013
No answers for : how to give work to two workers on the same Matlab session ?
Thanks !

Paul-Henri Michel
Paul-Henri Michel le 24 Avr 2013
Up :)

Catégories

En savoir plus sur Parallel Computing Toolbox dans Help Center et File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by