File talk:Mapreduce Overview.svg

This picture is not clear and misleading. The execution of "map" and "reduce" functions looks like inside the mastre node, but actually it is not. And this picture does not correctly describe the "parallel" characteristic of MapReduce: Data are reading in parallel and also output in parallel. Reading and writing does not through the master node. The master-node only tells slave-nodes which partition of input they should read.