MapReduce is the programming model that made processing petabytes of data on commodity hardware economically viable. Introduced by Google in 2004 and implemented as open-source Apache Hadoop MapReduce, it decomposes any large computation into two phases — Map and Reduce — that can be parallelised across thousands of machines independently. The model's power lies not in sophistication but in its extreme simplicity: if a programmer can express their computation as a map function and a reduce function, the framework handles all distribution, fault tolerance, and data movement automatically.
Although Spark has largely replaced MapReduce as the execution engine of choice for new big data workloads, MapReduce remains foundational knowledge for every distributed computing practitioner. Spark's RDD API is a direct evolution of the MapReduce model; understanding MapReduce's shuffle phase explains why certain Spark operations are expensive; and many legacy Hadoop pipelines still run MapReduce jobs in production. The mental model of map-shuffle-reduce also provides the clearest explanation of how distributed aggregations work at scale.