Don't be too clever
Thomas H. Cormen, Kristin Bruhl · 1995
We implemented and measured several BMMC permutations on the MasPar methods to perform MP-2.Our results indicate that, except for certain types of permutations or very high virtual processor ratios, the best method overall is the naive method but with virtual-processor numbers computed in Gray-code order.For some permutations, however, the naive method performs very poorly; the best method in these cases is an adaptation of the block BMMC algorithm for parallel disk systems in which the processor elements are treated as independent devices. 1