On a coarse grain MIMD machine, this algorithm is much more effective. In this case, we can label the clusters on the sub-lattice on each processor using a fast, efficient sequential algorithm. The local label propagation is now done only on the edges of the sub-lattices, to match up labels of clusters that cross processor boundaries.
Now have good load balancing, and the label of the largest cluster is
propagated across the number of processors, not the number of sites.
For an 
lattice on a 
processor array:
As long as 
, will get good efficiencies for this method.
