MATLAB

MATLAB is a high-level technical computing language and interactive environment for algorithm development, data visualization, data analysis, and numeric computation. More information can be found at: http://www.mathworks.com/

Note that as of MATLAB R2024a MathWorks has added a Service Manager to MATLAB startup that adversely affects startup time on the cluster. To keep that from happening, on the HPC3 please run matlab_fixes.sh from anywhere on the HPC3

MATLAB on the HPC3

Interactive MATLAB Sessions

Graphical on the FastX HPC3 Desktop:

qlogin -pe openmp 2
matlab -desktop

- or use Applications > Programming > MATLAB on the HPC3 Desktop (see FastX section here)
Textual:

qlogin -pe openmp 2
matlab -nodesktop

Submitting a MATLAB Job

Create MATLAB Commands File

Create a .m file with your commands, for example:

for i=1:1024
    A(i) = sin(i*2*pi/1024);
end

disp(A)

Create MATLAB Job Script

Create a .sh file with at least the following contents:

#!/bin/bash

matlab -nodisplay -nodesktop -nosplash < your-commands-file-1.m

or if your MATLAB command file calls functions in separate function files:

#!/bin/bash

matlab -nodisplay -nodesktop -nosplash < your-commands-file-1.m your-func-1.m your-func-2.m

Submit MATLAB Job

More information: HPC3 Job Management

Parallel MATLAB on the HPC3

Remember HPC3 Job Management with job limits for job queues.

Required to use parpool() only with an MPI cluster, run (on the command line, not in MATLAB): setup-parMatlab (this sets up your parallel environment in MATLAB). You only need to do this one time. After that, all MATLAB sessions will be ‘ready for parallel’.
To open a normal parallel MATLAB session, in your MATLAB, use the myPoolStart() function:

Usage of myPoolStart.m Function

The myPoolStart() function is in the /usr/local/demo/Matlab/parallel folder on all HPC3 systems. You can use the other files in that directory to build your own job scripts.

With MPI (multi-server) Workers

Start a master (could be MATLAB graphical or not):

qlogin -pe openmp 2 -l m_mem_free=2G

Start MATLAB and run the function:

# if you need > 2G RAM / worker:
export MATLAB_WORKER_RAM=3G
matlab -nodisplay
> addpath /usr/local/demo/Matlab/parallel
> myPoolStart("Wharton_HPC3_Matlab_R2023a", 16)
... do your parallel work!
> poolobj = gcp('nocreate'); 
> delete(poolobj)

With SMP (local) Workers

If you only need 8 or less workers (would be running on 4 cores, max), you should instead use the function like:

qlogin -pe openmp 4 -l m_mem_free=3G
matlab -nodisplay
> addpath /usr/local/demo/Matlab/parallel
> myPoolStart("Processes", 8)
... do your work!
> poolobj = gcp('nocreate');
> delete(poolobj)

“Processes” starts a lot faster. NOTE that RAM is per core, so the above example would request and require 12GB of RAM.

That function can be used in a job script as well.

MATLAB on my Laptop and Desktop

Wharton Site License!

Wharton subscribes to a site license for all Faculty, Staff, and Students. It expires annually on September 1st, after which you’ll need to re-authorize, and includes Base MATLAB, and a host of many excellent toolboxes.

To set up your Mathworks account and install MATLAB, please follow the directions HERE.