MPI acceleration with FEniCSx

[dokken]

I would suggest having a look at: Multiprocessing FEniCSx making simulation slower - #6 by dokken

which shows speed-up and the capabilities one can expect (for instance what size of problem is sensible).

It also refers to a previous posts:

Parallel slower runtime

I cannot reproduce your problem: root@82df035a5e65:~/shared# python3 mwe.py MESH CREATED 0 , time : 0.08017910199987455 root@82df035a5e65:~/shared# mpirun -n 2 python3 mwe.py MESH CREATED 0 , time : 0.0637212950000503 MESH CREATED 1 , time : 0.06375895700011824 root@82df035a5e65:~/shared# mpirun -n 4 python3 mwe.py MESH CREATED 0 , time : 0.06122965600025054 MESH CREATED 1 , time : 0.06122667699992235 MESH CREATED 2 , time : 0.061187144000086846 MESH CREATED 3 , time : 0.0612255729997…

Problem solving in parallel slower than in serial

This is due to garbage management in PETSc, see: Add explicit destruction of PETSc Python objects by garth-wells · Pull Request #2560 · FEniCS/dolfinx · GitHub which resolved this. Please note that the problem you are solving isn’t very large, i.e. num dofs= 27 783. In general I would suggest having between 50 000 - 100 000 dofs per process. Finally, I cannot reproduce your results with the docker image dolfinx/dolfinx:nightly: from ufl import TestFunction, TrialFunction, dx, inner, grad, …

which shows how much speedup you get in certain cases, and what part of the code gives optimial speedup.