MapReduce

MapReduce processes a large dataset in two stages across many workers. Map transforms independent input pieces into key-value pairs; Reduce combines all values that share a key. Keeping each task free of external side effects makes failed tasks easier to retry.