Matlab is currently available for use centrally on the HPC cluster systems of the University of Iowa. One may use the Matlab environment by loading the appropriate module:
[user@hpc:~]$ module load matlab_R2012b
While it is possible to run Matlab interactively, which can be especially useful for prototyping a job, please remember that login nodes are a shared resource and are intended for launching jobs or prototyping smaller versions of jobs you intend to run. It is not advisable to use interactive sessions for long-running, compute-intensive jobs unless one uses a qlogin session to do so.
Batch Matlab Jobs
A batch or serial job is generally run on a single node. The following is a very simple example of a non-parallel batch job using Matlab functions.
#!/bin/bash # The name of the job: #$ -N MatlabTest # Name of the output log file: #$ -o matjob.log # Combining output/error messages into one file: #$ -j y # Specifying the Queue #$ -q UI # One needs to tell the queue system to use the current directory as the working directory #$ -cwd # The command(s) to be executed: matlab -nodisplay -nodesktop -nojvm -r batch # Note after -r is the name of the routine or function
Here is the Matlab function you are calling from within your job script:
A = fix(100*rand(5,6)); B = fix(100*rand(6,3)); C = A * B
< M A T L A B (R) >
Copyright 1984-2012 The MathWorks, Inc.
R2012b (8.0.0.783) 64-bit (glnxa64)
August 22, 2012
To get started, type one of these: helpwin, helpdesk, or demo.
For product information, visit www.mathworks.com.
C =
12893 12980 15469
15263 19329 17603
7885 16894 17448
14434 20490 27329
9235 18735 22408
Parallel Matlab Jobs (or "pool jobs")
The current installation of Matlab uses the Parallel Toolbox, which allows for parallel jobs that can use up to 12 cores per job. Information on the Parallel Toolkit may be found here: http://www.mathworks.com/help/distcomp/index.html. The following example script can be used to submit a Matlab job to SGE to run on 8 cores (you may use all the cores in a node – which may be more than 8). Note that lines starting with '#$' are SGE shell commands, whereas '#' symbols denote comments, and the remaining lines are matlab commands.
There is some overhead to running jobs in parallel, so it can be slower to run on multiple cores if the job is small without sufficiently large loops. The easiest way to make a Matlab program parallel compatible is to use parfor (parallel 'for') loops. A parfor loop can be used when each iteration of the loop is independent of all other iterations. Here is a link to Matlab documentation on parfor loops: http://www.mathworks.com/help/distcomp/getting-started-with-parfor.html
Submit the job using 'qsub parallel-test.sh', assuming the name of the submission script is called parallel-test.sh.
#!/bin/bash # The name of the job: #$ -N test ## replace 'test' with job name # Name of the output log file: #$ -o matlabTest_parfor.log # Combining output/error messages into one file ( change y to n for separate files) #$ -j y # Specify the parallel environment (PE) and number of cores to use (8): #$ -pe smp 8 # One needs to tell the queue system to use the current directory as the working directory #$ -cwd # The matlab commands to be executed; replace "test" with your function name. # -r imediately runs a function without presenting an interactive prompt /opt/matlab/R2012b/bin/matlab -nodisplay -nodesktop -nosplash -r parafor-test
Within the Matlab script itself, you want to specify the number of cores you wish to reserve. This is done using the 'matlabpool('open', #)' command at the beginning of the file and ending the file with 'matlabpool('close')'. If you only want to use one core then you can omit the 'matlabpool' commands from your file.
matlabpool('open',8);
tic
start = tic;
clear A
parfor i = 1:100000;
A(i) = i;
end
stop = toc(start);
stop
matlabpool('close');