Problem solving in parallel slower than in serial

[dokken]

@cvillette

Sorry but I am not familiar with pull requests. Was this edit added into the latest dolfinx version? If not, how can I best use it?

Depending on how you create your singularity image, you should be able to pull the latest version of dolfinx/dolfinx:nightly and rebuild your image on that.

@cvillette

I have tried using the ‘hypre’ pre-conditioner but performance did not increase (actually both total running time and relative gain in time with increasing number of cores were slightly worse). Would the benefit only appear for higher number of dofs? Also, would you suggest another pre-conditioner?

I am not an expert in pre-conditioning, so I will leave this questions for others to answer. You could have a look at: Not getting a better performance in parallel - #2 by dokken