Parallel Swendsen-Wang cluster algorithms for moderate sized lattices
work fairly efficiently for moderate numbers of nodes 
on a MIMD machine. However, this does not prevent us from using
massively parallel MIMD machines with 
nodes for this problem.
Can use a hybrid of domain (data) parallelism and independent parallelism (over different MC runs). For example, can run 8 independent simulations of 64 nodes each on a 512-node machine.
This gives much better performance than running parallel cluster algorithms on vector supercomputers, since these applications do not vectorize well.