I am using a METIS graph partitioning for domain decomposition of my unstructured code, but even for very simple topology METIS returns a non-optimal decomposition. For example for a simple 8x8x8 cube mesh with 8 partitions I'd expect 8 4x4x4 cubes but I get something non-"symmetric". I've tried messing with the node and edge weighting without success. Anybody have any ideas? Thanks Rob

I use METIS too, to partition unstructured meshes. I have noticed the same property of METIS. When i supplied a very small 2D mesh of triangles, METIS produced a non-optimal partition into 4 pieces.

You might contact the authors of the METIS codes to see if it is possible to modify the algorithm to give an improved (not necessarily optimal) answer in these limitingly small mesh cases. However, on any (larger) mesh that i would actually use for a flow computation, METIS certainly produces a good partition. Also, it is extremely rapid.