MapReduce is a programming model for parallel and distributed processing of data. It consists of two steps: Map and Reduce. These steps are inspired from functional programming, a branch of computer science that deals with mathematical functions as computational units. Properties of functions such as immutability and statelessness are attractive for parallel and distributed processing. They provide a high degree of parallelism and fault tolerance at lower costs and semantic complexity.

In this chapter, we will look at advanced optimizations when running MapReduce jobs on Hadoop clusters. Every MapReduce job has input data and a Map task per split of this data. The Map task calls a map function repeatedly on every record, represented as a key-value pair. The map is a function that transforms data from one domain to another. The intermediate output records of each Map task are shuffled and sorted before transferring it downstream to the Reduce tasks. Intermediate...