Distribution Mapping that minimises cost for syncing AMR patches.

[ankithadas]

Is there a way to make a distribution mapping that is aware of the coarse AMR patches such that syncing AMR patches (average down) minimises inter-processor parallel copy ? I would like to develop a mapping that eliminates inter-processor parallel copy altogether. Any help would be greatly appreciated!

[WeiqunZhang]

#4634 (comment)

[ankithadas]

I understand it can cause performance issues. It would be nice to know how I could minimise inter-process parallel copy (that is fine and coarse boxes belonging to the same MPI rank are maximised) based on some cost function. Ideally, the distribution mapping is made based on both communication costs and work costs. Is that possible ?