MPI tutorial compatible with Fenicsx 0.7+

General
1

Hi,

I’m looking to understand how to parallelise my project in fenicsX with the supported openMPI framework. I was looking for resources for this and could only find this.

Unfortunately, as soon as fenicsx calls are used the example scripts do not run - I presume the API has changed in more recent versions.

Are there any up to date resources on this?

Thanks,

R.

2

You could check out @dokken’s tutorial: The FEniCSx tutorial — FEniCSx tutorial

If you go to the source you can check out and build the tutorial at the specific verison you want, e.g., v0.7.0:

There might be a way to get this to work online without having to build it yourself; however, this is beyond my current knowledge.

I think @dokken also has some tutorials with emphasis on MPI, but I’ll let him relay the links.

3

Here is the DOLFINx + MPI tutorial for v0.7.x
https://scientificcomputing.github.io/mpi-tutorial/notebooks/dolfinx_MPI_tutorial.html
written by me and @hherlyng. It is inspired by the NewFrac code.

1 Like
4

Great, than you.

Just quickly, before I wrestle with that, my project is using the non-linear assembly with dolfinx-MPC to get periodic boundaries, as implemented here and discussed here. Will application to that be the same? Or will that create any significant hurdles?

Thanks,

R.

5

Not sure what you refer to here?

DOLFINx_MPC is built as an extension of DOLFINx, and uses DOLFINx structures, such as dofmaps, index-maps, functions etc.

The only major difference between the DOLFINx and DOLFINx_MPC is that DOLFINx MPC creates special index_maps and dofmaps with constraint information, and does assembly in a specific way, as specified in https://fenics2021.com/slides/dokken.pdf

6

Sorry, I’m not particularly fluent with all of the machinery and am guilty of “cargo cult coding” my way through to some extent.

So I was just wondering if the application in the link you provided - which implements a linear solver without the MPC extension - can be applied to my case, in terms of dolfinx calls, without further guidance/modification of the function calls etc. So the hurdles I was anticipating were really my adapt it to my case, not technical issues with MPC etc.

I’ll give it a go first and see if I run into any trouble.

Thanks.