
310 Big Data Computing
MapReduce is able to start running jobs as soon as the required data are
available in the le system. First of all, jobs are initialized by creating a series
of map and reduce tasks. The number of map tasks is usually determined
by the number of splits into which the input data are divided. Splitting the
input is what makes it possible to parallelize the map phase and can have a
great impact on the performance, so splits can also be thought of as the rst
level of granularity of the system, and it also shows how the le system and
MapReduce are integrated. For instance, if the input consists of a single 6.25
GB le ...