CanChen [email protected]
MapReduce
The MapReduce model in user program first split up the input file into M parts and then forks user program to every node in the cluster. One node is special and called master. Master node assigns M map tasks to nodes and these nodes process the corresponding data part and write the results in the local disk. Here written data is splitted up into R parts and after that the node which receives reduce task will read one part of the buffered data remotely. The results of R tasks are then written out as the final result. For me I think the R results should be reduced to the real final result.