-
Notifications
You must be signed in to change notification settings - Fork 1
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add Dockerfile for easy deployment on NVIDIA systems #34
Comments
But Dockerfiles won't help with MPI environments - MPI libraries depend on the lower-level libraries installed on specific HPCs, that are configured for local hard- and firm-ware. |
You need to install |
I don't think this will work for a general case, as you need MPI libraries built first, and configuration/installation of those is hardware-dependent. I hence don't believe there exists a containerised solution to at least MPI-based (and maybe any distributed) calculations. However, @NathanCQC indeed used https://docs.nersc.gov/development/shifter/ on Perlmutter, but I am not sure it will work for arbitrary multi-node system. I am happy to be wrong, though. |
We could do this via the CI. But there is one already built somewhere. But it is specific for Perlmutter. as they have special MPICH. IMO containerisation on HPC is still very system specific, although I think it is improving. and is not true to the docker ideas about portability etc, Did you have in mind a specific hardware provider? |
I had in mind Perlmutter at the moment. |
@NathanCQC what do you mean by doing this via the CI? |
I think this is for @NathanCQC
It would be good to have a Dockerfile that simplifies the deployment on NVIDIA systems, in particular considering parallel MPI environments (multi-node and multi-GPU).
The text was updated successfully, but these errors were encountered: