Body
With regard to using MATLAB in computing clusters, HPC centers, clouds, and related scheduled environments, please note the following.
There’s a specific product in our Campus Wide License that is specially designed for use in high performance computing environments: MATLAB Parallel Server (formerly called MATLAB Distributed Computing Server or MDCS)
NOTE: This product is on its own License Number, separate from the main Campus Wide License: 40710858.
Mathworks recommends adding any of our HPC admins to the license due the unique requirements for installation and configuration.
The end-user workflow is vastly different when you use MATLAB Parallel Server (ML PS) vs. invoking an arbitrarily large number of MATLAB and Parallel Computing Toolbox (PCT) instances:
- MATLAB and PCT would require the user to mange integration with our scheduler, write the code that distributes the workload and does load balancing as well as concatenating the results
- MATLAB and PCT are also somewhat limited in that each MATLAB instance will only take advantage of the cores within the physical ‘computer’ (blade, system, motherboard) on which it is running
- MATLAB Parallel Server only needs the end-user to make some boilerplate updates to the same code they’re running on their local system using PCT, and select the cluster’s name from a dropdown menu to submit the job.
- ML PS has no limits to the number of cores it can be used on, nor the number or campus compute clusters. (NOTE: This was not the case prior to 2019b release, so this is likely new information for many HPC Admins)
- MathWorks support will work with our cluster admins on the scheduler integration and the creation of a “profile” which makes this workflow possible.
This generally encourages “non-programmers by trade” to utilize HPC resources they might otherwise need to enlist a computer scientist to use.
Both use cases are included with our Campus Wide License, but using ML PS may broaden the cluster use base due to the simplified workflow.
You’ll find further information on MATLAB Parallel Server here: https://www.mathworks.com/products/matlab-parallel-server.html
(As noted above – licensing terms for Campus Wide License customers does NOT include any limit to the number of instances/workers) that can be invoked.)