I agree with you there. What I am saying is that is there anyway to take the data that you have in the map phase and group it in a manner such that the reduce phase maybe runs more times but is processing less data chunks