Running ANSYS from Docker container

[chart-magnum]

Hello!

First of all, congratulations on a very useful and powerful project!

I'm writing with a question concerning the use of ANSYS with Docker. I perform an electromagnetic simulation with an in-house code that runs on CENTOS with jupyter notebooks. In addition, my team is using ANSYS installed on Windows. I need to run together (in-house code and ANSYS) in a scripted way. To this end I have created a Docker image and execute the in-house code on Windows via jupyter notebook exposed by Docker. My question is: what would be an advised way to call ANSYS in this setup? Should I add ANSYS to the docker image, or would it be possible to call ANSYS MAPDL on Windows from jupyter notebook on Docker?

Thank you in advance for any hint you may offer.

Best regards, CM.

[akaszynski]

I'm writing with a question concerning the use of ANSYS with Docker. I perform an electromagnetic simulation with an in-house code that runs on CENTOS with jupyter notebooks. In addition, my team is using ANSYS installed on Windows. I need to run together (in-house code and ANSYS) in a scripted way. To this end I have created a Docker image and execute the in-house code on Windows via jupyter notebook exposed by Docker. My question is: what would be an advised way to call ANSYS in this setup? Should I add ANSYS to the docker image, or would it be possible to call ANSYS MAPDL on Windows from jupyter notebook on Docker?

Great question!

There's two ways to do this as your pointed out:

• Package MAPDL + Jupyterlab within a single container and call it as a single unit. The advantage of this approach is it's quite portable (you can move this to other computers/OSes) and ensure you have all the required dependencies. Disadvantage is size of the image and any potential performance hits or GPU issues (GPU support for tensorflow is difficult to achieve on Windows)
• Ansys on host, jupyterlab on docker image. Advantage of this approach is you'll see better performance running natively on the host. I've not benchmarked it myself (and this will likely depend on the type of solution), but you'll probably see better performance (especially if you're not running WSL2).

what would be an advised way to call ANSYS in this setup? Should I add ANSYS to the docker image, or would it be possible to call ANSYS MAPDL on Windows from jupyter notebook on Docker?

If it's not too much effort, I'd try both. To get the Ansys on host, jupyterlab on docker case working, run MAPDL in server mode -grpc and then expose that port (default 50052, you'll see it in stdout) and then talk to that through your jupyterlab instance.

Curious about your use case, did you see the docker image we've posted here?
https://github.com/orgs/pyansys/packages?repo_name=pymapdl

Also, would you mind emailing me at my work address at [email protected]? I'd like to get in contact with you regarding containerization strategies. If Ansys can provide something useful for your aims, I'd like to make that happen.