Distributed AMR
Speedup factor for special cases of a number of sub-grids
spread uniformly in the coarse mesh as a function of the number of processors.
The special case here is that no re-refinement is done, and hence this represents
an ideal problem for distributing AMR. In red are the results for
8 grids (2 per side), in green are the results for 64 grids, and
blue shows results for 125 grids (5 per side). The purple
shows the ideal case. Performance saturates roughly when the number of processors reaches the half the number
of grids.
Distributed computing via domain decomposition
Adaptive Mesh Refinement (AMR) (serial) in a flatspace 3d nonlinear model
Adaptive Mesh Refinement (AMR) (serial) in a gravitating 2d collapse model