Wolfgang, Thanks for your response. That makes a lot of sense. I suppose they are many competing requirements in defining an "optimal" partitioning, particularly as you go to large numbers of processors.
> To make this less likely, deal.II sorts the coarse mesh cells by a > Cuthill-McKee algorithm, which is why you see the concentric shells: Cuthill- > Mckee orders the cells one shell at a time. > > I think this algorithm makes things a bit better, but it's certainly still > not > good. I think it's also not bad once you have many more processors than > coarse > mesh cells -- or, alternatively, if you have many more coarse mesh cells than > processors. The only case that's problematic is when the number of processors > is roughly equal or slightly less than the number of coarse mesh cells, which > I think is the case you run into. > > If you have ideas how to improve this all, I'd be interested in working on > this with you! I wouldn't even describe my case as problematic. The partitioning works fine, it just struck me as unusual. My only thought would be to pre-partition the coarse mesh using, e.g., METIS, and then perform a Cuthill-McKee ordering within the partitions. Are there any continuity requirements for the space filling curve, or can we rearrange coarse cells arbitrarily? Like I said though, the current strategy doesn't seem to be a problem, so this may not be worth the additional effort. > > Best > W. > > ------------------------------------------------------------------------- > Wolfgang Bangerth email: [email protected] > www: http://www.math.tamu.edu/~bangerth/ _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
