How to run a job on a remote cluster?

16 vues (au cours des 30 derniers jours)
Ryan
Ryan le 4 Oct 2014
Commenté : Aleksander Lidtke le 24 Août 2015
I created a bash script that I have uploaded to a remote cluster via scp. In this script I load a Matlab module (version R2014a) which I use to run some Matlab code.
#$ -pe smp 4 //tells the cluster to provide 4 cores local to a single node
module load matlab14/R2014a
matlab -nosplash -nodisplay -r "code I want to run"
Within my Matlab code I am creating a parallel object and running the following (I'm trying to learn how to run a job in parallel on the cluster before writing the real code).
pool = parpool('local',4);
parfor k = 1 : 10000
disp(k); //console input saved is to file to demonstrate I have the general syntax right
end
delete(pool);
When I run my code I get the following error
{^HUndefined variable "com" or class
"com.mathworks.toolbox.distcomp.pmode.SessionInfo.NULL_SESSION_INFO".
Error in parpool (line 96)
sessionInfo =
com.mathworks.toolbox.distcomp.pmode.SessionInfo.NULL_SESSION_INFO;
Error in main (line 4)
pool = parpool('local',4);
Error in run (line 63)
evalin('caller', [script ';']);
}^H
^G{^HError: File: /Users/name/scripts/main.m Line: 4 Column: 1
The input character is not valid in MATLAB statements or expressions.
Error in run (line 63)
evalin('caller', [script ';']);
}^H
^G{^HUndefined variable "com" or class
"com.mathworks.toolbox.distcomp.pmode.SessionInfo.NULL_SESSION_INFO".
Error in parpool (line 96)
sessionInfo =
com.mathworks.toolbox.distcomp.pmode.SessionInfo.NULL_SESSION_INFO;
Error in main (line 4)
pool = parpool('local',4);
Error in run (line 63)
evalin('caller', [script ';']);
}^H
^G
What might I be doing wrong and how might I fix it? I have been able to run other Matlab scripts but I have not been able to run any that implement parallelization. I checked the cluster and the Parallel Computing Toolbox is present.
  1 commentaire
Aleksander Lidtke
Aleksander Lidtke le 24 Août 2015
Hi, I just ran into the same issue. My supercomputer help page suggests starting Matlab with -nodisplay -nojvm -nosplash options. It turns out that -nojvm causes this error. The variable com the error message mentions is, in fact, some Java module, so it won't work without the Jave Virtual Machine.

Connectez-vous pour commenter.

Réponses (1)

Edric Ellis
Edric Ellis le 6 Oct 2014
Parallel Computing Toolbox relies on Java, so you should omit the -nojvm flag from your matlab command-line.
  2 commentaires
Ryan
Ryan le 6 Oct 2014
Yes, you are correct. I had already removed it and forgot to remove it for this post. It did not change the results I was getting though. I have updated my post.
Edric Ellis
Edric Ellis le 6 Oct 2014
Hm, the error message that you get (unhelpful though it is) is precisely the error that you see when specifying -nojvm, and I know of no other way to get that error. Perhaps you could try adding
error(javachk('jvm'))
to the start of your script - that will error out if Java is not enabled before you get any further.

Connectez-vous pour commenter.

Catégories

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

Produits

Community Treasure Hunt

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

Start Hunting!

Translated by