Installing Matlab

Matlab currently needs to be installed manually, as the Spack package only offers very old versions. This install is also annoying because it uses a GUI installer. Using a virtual desktop like FastX or VNC is recommended.

To install Matlab, you will need to first download the Linux installer from their website either in the remote desktop, or locally and then upload it. You need to have a Mathworks account to do this.

Matlab Download Page

I have typically stored these downloads in /glade/work/csgteam/build/matlab. Once downloaded, unzip into a version-specific directory and run the installer script ./install.

Tip

As you will need to run the installer as csgteam, you should ensure that you sudo to csgteam using X11 forwarding. You can do this with sudox csgteam.

The installer will also require you to log into a Mathworks account tied to our license. It will then ask you to select a license to use; probably you will want to select Matlab Parallel Server, as this will ensure the MPS is installed along with other toolboxes. At some point, it may ask if you want to provide a license file. You can point it to the network.lic file in the build directory mentioned above.

It will then ask for a path - put it in /glade/u/apps/opt/matlab/<version>.

Turn off the "Improve..." checkbox, confirm and then expect it to run for 20-60 minutes.

Adding to Spack

Once Matlab is installed, it can be added to Spack as an external via the constraints.cfg file and then "installed" into the cluster. By "installing" it in Spack, we motivate Spack to generate a module for that Matlab version.

Cluster definitions should already contain important module customizations for Matlab. In particular, ensure that Matlab modules contain the following configuration:

matlab:
  environment:
    set:
      MW_FEATURE_MathworksServiceHostIsOnByDefault: '0'

This undocumented environment variable will tell Matlab (at least 2024-2025 versions) to not launch the ServiceHost background process. This process has no tangible benefit for our users and in fact downloads ~1 GB of data into their home directory.