Matlab

The majority of Matlab users at the Hutch have Matlab installed on their desktop system. Most users have dedicated licenses for their desktop installs Matlab. Dedicated licenses offer the flexibility to use Matlab with a laptop on or off campus. The center has a limited number of shared network licenses. Users are allowed to use a maximum of 4 concurrent Matlab licenses from the license manager. The shared network licenses can be used with the centers Linux cluster. This document describes how to use Matlab on the Gizmo cluster.

Interactive Matlab on Linux

Before you Start

If you want to run Matlab interactively on one of SciComp’s compute nodes you will need to forward the display from that remote system to your desktop. noMachine is the recommended way to do this.

NOTE: Do not run computationally intensive processes on the noMachine systems or the rhino nodes. Use these only for development, design, and testing. Once your process is designed and tested, plase use the grabnode command to request and allocate a cluster node for your process.

From a terminal in a noMachine session:

• ssh rhino
• Type grabnode, then answer the questions about how many CPUs and Memory you require. When your grabnode is successful you will have a Linux command prompt on a gizmo cluster node.
• module load MATLAB/R2022B
• Type matlab and press enter. The Matlab window should open in your Linux session. Success!

Running Matlab Batch Jobs on a Cluster

Matlab programs can be run on the gizmo cluster using sbatch. You can read more about using sbatch in general on our Job Management page. This allows you to use additional resources that are not available on a desktop install. Note that the Matlab program and data need to be available to the Gizmo/Rhino systems. This is typically accomplished by copying these files to your Fast directory.

When you have the data in place, start a session on a rhino using ssh or NoMachine). Then create a batch script that will request resources on gizmo and then run your Matlab program like the following:

#!/bin/bash
#
#SBATCH --qos=matlab
#SBATCH --timelimit=3-0
#
# These two lines need to be in your script to set up the Matlab environment
. /app/lmod/lmod/init/profile
module load MATLAB/R2022B

matlab -nodisplay -nosplash -nodesktop -r  "run('myAnalysisJob.m $1 $2'); exit;"

The exact options above can be altered depending on your computational needs- contact Scientific Computing if you need assistance determining the needs of your script. However, the option --qos=matlab is necessary for license management.

MATLAB Distributed Compute Engine

The distributed compute engine allows you to run parallelized Matlab programs across multiple cores and/or nodes simultaneously. Tutorial and setup guide for the parallel tool box is in the wiki-code-examples repo.