The amount of parallelism depends on the size of input.We have Master and worker nodes.Few workers perform Map functionality and few perform Reduce functionality.Both the Map and Reduce are user supplied functions.Map parses the input given and produces a Map(Key,value).Reduce performs the required functionality picking the list of values reduce(ki,list(vi)).This systems is fault tolerant when the workers fail but the entire map reduce task needs to be aborted if the master fails.
