HPC University of Fribourg

Trace: » matlab_-parallel_version_with_parallel_toolbox


|

Meta

Matlab R2010a is installed on the cluster

Please use queuing system and submit batch job or if you want to use graphical interface ssh to empty node (qstat -f to check load of the nodes 0 means empty 8 means full). Never submit job directly on beo03, use nodes instead.

/apps/bin/matlab

You can use up to 8 CPUs on each node


Recommended method to submit MATLAB in graphical mode:

ssh yourusername@beo03.unifr.ch

1CPU usage:

qrsh xterm

/apps/bin/matlab

8CPUS usage:

qrsh -pe smp 8 xterm

/apps/bin/matlab


We have 10 Matlab licenses for the cluster and 10 parallel toolboxes. If you need another toolbox we need to order it. Use parfor to parallelize your jobs.

By default Matlab, makes use of the multithreading capabilities of the cluster. This means that when Matlab is running on a node of the cluster it uses very often the 8 cpus of the node, even if the option “ #–pe smp 1 ” is used. To force Matlab to use only on cpu of the node, it has to be started with the following option: -singleCompThread. Alternatively in the Matlab code, it is possible to set the number of maximum cpu that Matlab can use with the command: maxNumCompThreads(N)


To submit matlab in batch copy runmatlab from /apps/bin/ folder

Than submit it with qsub

qsub runmatlab input_name

Remember that input_name is without .m extension.

Script is for 1 CPU jobs.