Home » Forums » METIS » Graph & Mesh Partitioning

Weighting limits for decomposition

Dear All,
we are using metis to decompose octree-type data structures for CFD computations. Although the level difference of neighbouring blocks is restricted to 1 (meaning that a block face has a maximum of four neighbours with half the size each our numerical scheme requires a weight difference of sixteen between blocks differing by one level. When using 5 grid levels, this translates into wight ratios of 16^4 and metis works fine. If we increase to 6 levels (implying a weight difference of 16^5>1000000), the decompositions are completely unsatisfactory. I would be grateful for any hint regarding limits of weight ratios in metis and how to avoid them.
Thanks & best regards
Manfred Krafczyk

‹ Is there a way to specify a maximum size to Metis or hMetis Forcing two elements to be on the same partition ›
Submitted by manfred_krafczyk on Tue, 2011-03-22 09:42
»

RE: I'm not sure that I

I'm not sure that I understand the question. Is this a case in which you want the partitions to be balanced and they are not, or is it a case in which you want the partitions to be unbalanced?

Submitted by karypis on Wed, 2011-03-23 12:29.
»