Gridding and Load Balancing
MFIX-Exa has a great deal of flexibility when it comes to how to decompose the computational domain into individual rectangular grids, and how to distribute those grids to MPI ranks. There can be grids of different sizes, more than one grid per MPI rank, and different strategies for distributing the grids to MPI ranks.
We use the phrase “load balancing” here to refer to the combined process of grid creation (and re-creation when regridding) and distribution of grids to MPI ranks.
See Grid Creation for how grids are created, i.e. how the BoxArray on which
MultiFabs will be built is defined at each level.
See Load Balancing for the strategies AMReX supports for distributing
grids to MPI ranks, i.e. defining the DistributionMapping with which
MultiFabs at that level will be built.
MFIX-Exa also allows for the “dual grid approach”, in which mesh and particle data are allocated
on different box layouts with different mappings to MPI ranks. This option is enabled
by setting mfix.dual_grid = 1 in the inputs file.
See Dual Grid Approach for more about this approach.
When running on multicore machines with OpenMP, we can also control the distribution of
work by setting the size of grid tiles (by defining fabarray.mfiter_tile_size), and if relevant, of
particle tiles (by defining particle.tile_size). We can also specify the strategy for assigning
tiles to OpenMP threads. See
MFIter with Tiling
in the AMReX documentation for more about tiling.