How to run fmincon on a cluster?

Hi all, I want to run a matlab function, containing "fmincon" on a cluster. I used this code on my PC to use all of 4 cores.

matlabpool open 4
options=optimset('UseParallel','always'); 
[u,fval,exitflag] =fmincon(J,u0,[],[],[],[],lb,ub,[],options);
matlabpool close

When I want to run the same code on a cluster it returns this error: Error using ==> matlabpool>iVerifyJava at 379 matlabpool is not supported because Java is not currently available.

Error in ==> matlabpool at 92

Answer 1

[Edric Ellis]

How are you running the code on a cluster? It looks like you may be running "matlab -nojvm" on the cluster - as the error message suggests, you need the JVM available to run "matlabpool", so you need to run "matlab -nodisplay" instead.

Rather than using standard MATLAB/PCT licences on the cluster, you may wish to consider using MATLAB Distributed Computing Server licences ( MDCS product page ). That way, you can submit a "matlabpool job" directly from your desktop to the cluster where it will execute consuming only MDCS licences.

Answer 2

[Victor]

Hi all, Actually some times it depends from one cluster to another cluster, but you can try this steps: (Good luck :))

No matter which version you are using, the number of threads can be significantly reduced by turning off MATLAB's Java Virtual Machine (JVM), and forcing it to use a single computation thread. This can be done using the options:

matlab -nodisplay -nojvm -singleCompThread

(Note: -singleCompThread is supported only in MATLAB 7.8 and 7.9.)

If your MATLAB program is using a multithreaded library such as Intel MKL, you should also set the number of threads to one before executing "bsub":

export OMP_NUM_THREADS=1

bsub -n 1 -W HH:MM [...] matlab -nodisplay -nojvm -singleCompThread [...]

If, for any reason, you really want to use multiple threads, you must request an equivalent number of CPUs when you submit your job:

export OMP_NUM_THREADS=N

bsub -n N ...