How MapReduce Works
There is a single Master computer and many Worker computers.
The Master divides the input data into bite-size chunks of 64 MB and assigns the data chunks to workers. If possible, Master chooses a worker that already has the data on its hard drive in the Google File System, or is close to a computer with the data; this minimizes network traffic.
Think of the data chunks as being like a sack of beans: lots of pieces of data, all more or less alike.