Open kktach opened 4 years ago
I did some poking around and didn't find anything specific that I could see that would cause this. Does this only occur in this split mesh test? Can you check if this occurs on the master branch for the exact same test?
Is the current version of the test pushed? I didn't see the setting of the weights, but its been awhile since I looked at EnGPar code so I could have just missed it.
I've replicated the test on master and the same behavior was found there. Digging in, the growth of part boundary (sideTol
) is being triggered which is preventing further balancing. This is mostly due to the heavy initial imbalance and only using two processes.
I added an exception in the above commit to master that ignores the side tolerance on two processes.
On branch kk_hg_bfs. Mesh is not in pumi-meshes repo but can be copied from /lore/cwsmith/geometries/pumi-meshes/upright/1.6M/2p. Verbosity in engpar::balance was changed to 2.
There are weights (1.7) applied to the mesh entities in process 1, which is used to imbalance the mesh during splitting. The diffusive balancer stops after 1 step. Output says no vertices migrated so it stopped. Why does this happen?