Non-interactive installation on shared storage

I need to install Matlab on a number of Linux/RedHat9 VMs. Matlab will be used by multiple users on these VMs. All VMs share the same NFS/GPFS storage for user accounts. I would like to automate the installation using Ansible and Matlab's "silent" installation feature. A simple method would be to install Matlab to shared storage where all users are able to see it. This would require only a single installation for multiple machines, with a significant reduction in overall disk space required. However my attempts so far suggest that some parts of Matlab are not able to work on NFS file systems because NFS cannot reliably lock files. Is an NFS installation possible (without having to write a complex and potentially fragile wrapper script to set environment variables that control where Matlab stores resources that require file locking)?

Answer 1

[Isha]

Hello,

MATLAB supports silent (non-interactive) installation and centrally managed network licensing, so it can be deployed at scale using automation tools and a hosted license manager.

Concurrent License Administration - MATLAB & Simulink

Install Products Noninteractively - MATLAB & Simulink

A shared install on NFS is possible, but MATLAB still creates per-user and per-host runtime files in writable locations such as the user profile. So even with shared binaries, some components remain local. [mathworks.com] [mathworks.com]

.MathWorks/ServiceHost directory fills up for NFS-Installed MATLAB - MATLAB Answers - MATLAB Central

NFS can also introduce file-locking issues, which may cause hangs or inconsistent behavior in distributed setups.

For that reason, a fully shared NFS-only setup is generally not reliable. A hybrid model is usually best: keep MATLAB binaries on shared storage, but use local per-user or per-node space for temp files, preferences, and caches.

Hope this helps.

[Simon]

Thanks Isha,

I've got a bit further with my understanding of Matlab installation. I was looking for something like dnf/yum installs on Linux, which in the main can be completely non-interactive from the start. If I understand the Matlab process correctly, you must download the installer, then use it with your license to add Matlab packages to an installer folder on your local machine. All this must be done interactively, and only at that point can you run the "silent" installer non-interactively.

My confusion with NFS was about the separation between the (presumably) read-only executable/data parts of the software and the "per-user" and "per-host" runtime parts. With Matlab 2026a we are getting errors like these on startup:

[3642680:3642680:20260511,173028.963030:ERROR file_io_posix.cc:226] flock: Remote I/O error (121)

They relate to files in the users ~/.Mathworks folder. A support request identified that these are, in the main warnings, and can be ignored. However it took me some effort to identify which files were causing the message and to understand the difference between these runtime files and the core Matlab code, and to realise that our installation was, in fact, working OK.

Our systems are set up in such a way that users generally don't have access to local storage (apart from places like /tmp) - all their storage is on shared space. So the option of using local space for runtime files doesn't exist without a major change in how we create our VMs. I would imagine that this type of setup is not unusual in an enterprise envrionment where Linux is heavily used, or in HPC clusters.