Matlab

Simple examples for running Matlab tasks can be found in the /home/SOFT/modules/examples/Matlab/ directory, which you can copy to your own using:

cp -r /home/SOFT/modules/examples/Matlab/ ./

In the single directory there are two files, the Matlab script “montecarlo_single.m” and the script for sbatch “slurm.sh”. By typing

cd single
sbatch slurm.sh

will start a program on the queue system on a single core that takes about 20-40s to run (kraken-m1,…-m6 machines are faster on this job).

There is a similar example in the “parallel” directory for a job parallelized over 8 cores within a single node. The script for sbatch is:

#!/bin/bash
#SBATCH -J example_matlab
#SBATCH -N 1
#SBATCH -n 8
#SBATCH --time=2:00
#SBATCH -o slurm.%N.%J.%u.out # STDOUT
#SBATCH -e slurm.%N.%J.%u.err # STDERR
matlab -nodisplay -r "montecarlo_parallel;quit;"

Running a Matlab job across multiple nodes is not possible because the “Matlab Parallel Server” is not purchased. Therefore, you must always keep the option in the startup script

#SBATCH -N 1

Parallel methods in Matlab programs are initialized with the “parpool” function. In our setup, a program using 8 cores must contain

parpool('local', 8);

However, automaticall loading of the number of cores allocated by Slurm is possible and recommended for beginers. Slurm stores the value defined by “#SBATCH -n” in the system variable “SLURM_NTASKS”, which Matlab can read

sz = str2num(getenv('SLURM_NTASKS'));
parpool('local',sz);