METIS typically works well and is exceptionally fast on a single core with a large numbers of cells. Unfortunately, it can generate zero-cell partitions. We need to find a partitioner that is both fast and robust or at least allows for a rebalancing step. SCOTCH is too slow for meshes with lots of cells. It may also be necessary to do our partitioning outside of PETSc. In any case, this is an area that needs exploration and testing.
METIS typically works well and is exceptionally fast on a single core with a large numbers of cells. Unfortunately, it can generate zero-cell partitions. We need to find a partitioner that is both fast and robust or at least allows for a rebalancing step. SCOTCH is too slow for meshes with lots of cells. It may also be necessary to do our partitioning outside of PETSc. In any case, this is an area that needs exploration and testing.