The idea behind cluster MC algorithms is very simple. Local MC algorithms
perform poorly because they update only one spin at a time. Cluster
algorithms use a clever way of finding large clusters of sites which can be
updated at once. (Remember that trying to update a group of sites chosen at
random gives a large 
and thus a small
acceptance).
Cluster algorithms were invented by Swendsen and Wang in 1987, for Ising and Potts spin models. They have since been generalized to other models, and other algorithms.