How to use Matlab Parallel Server

Hello,

We are using Simulink for running some heavy simulations. Our university has a licence for Matlab Parallel Server. And here our struggles begin...

How can we use the Matlab parallel server? The uni is saying that I can assign the license in the license center, but I neither see it, not can I enter it. Do they have to do it?

Once I have this server what are the next steps? Is it supposed that the uni has a cluster, or I can create one on my own using the comptuters I have?

I have read a lot of the online documentation of the Parallel Server and I still cannot fully understand the use cases and requirements.

I would be really helpful if someone can answer these questions.

[Jason Ross]

A number of these questions can only be answered by the staff at your university -- the license questions are related to what kind of license your university has, and if they have a cluster they likely have instructions on how to access it, how to submit jobs, etc.

To access and submit jobs to an existing cluster, your client only needs the Parallel Computing Toolbox license (use "ver" at the MATLAB command prompt). So if you have that and can find documentation for how to connect to a cluster, you should be able to submit jobs.

As for cluster creation, you can set up a standalone cluster using Parallel Server or integrate with an existing cluster. But that's a lot of work/effort compared to connecting to an existing cluster that's already set up, so it is a good idea to find out if there is existing documentation. Depending on how your university publishes documents you may be able to just google it with searches like "submitting matlab jobs to <university> cluster" or "using parallel server at <university>". These searches work for a number of universities I've tried.

While you are figuring out the cluster question, you can actually experiment on your local machine using the "local" cluster profile, which should require no setup and will be able to start workers, etc just like the cluster. When you figure out the cluster access questions, you can change the cluster profile to use the one for your compute cluster vs. local machine. So you can get experimenting right now.

[Iliya Valchev]

Thank you for your answer. I was happily surprised that these searches brought some results; great support!

I am still not sure I understand correctly the concept of the Parallel Server. Is it correct that the orgranization needs a license for it from Mathworks, and then configure on its own this server and provide its own computational power? So Mathworks does not provide the server as a service, and also just installing the Parallel Server is not enough and you need to manually configure some scheduler and computational cluster?

[Jason Ross]

We support a number of environments, including on-premesis and a service in the cloud. In my experience, universities often have on-premesis HPC/compute facilities for research computing that is supported/configured by university staff. In these scenarios Parallel Server integrates with an existing job scheduler (Slurm, LSF, Torque, PBS, etc) so that MATLAB jobs can run alongside other parallel compute jobs.

If you just want to install locally on some hardware that you have, you can install Parallel Server and use the job scheduler provided.

If you don't want to deal with the setup and maintenance for the server and hardware/etc you can run a cluster in the cloud. There is an option for us to do all the setup, all you need to bring is your AWS or Azure credentials (Cloud Center), or we publish reference architectures if you prefer to build your own cluster on the cloud. We also work with other hosting providers who will host your cluster if you don't want to use AWS or Azure, and don't want to build it yourself (see the link for those providers).

There are pros and cons to all the above in terms of cost, difficultly/amount of work, level of effort, etc.