He did a decent job of explaining how the different parts of the process relate to each other when it comes to the real world situation of parallelizing MR. Th

He did a decent job of explaining how the different parts of the process relate to each other when it comes to the real world situation of parallelizing MR. The shuffle (grouper) actually happens per reducer in many implementations using hash-bucketed outputs from the mappers, but that's an optimization.