Spark is a MapReduce-like cluster computing framework designed for low-latency iterative jobs and interactive use from an interpreter (think about sequences of map, filter, reduce). It can load data from any Hadoop InputFormat, and can run on Apache Mesos or in standalone mode.