Resilient distributed datasets, more commonly known as RDDs, are the primary data structure used in Spark. RDDs are essentially a collection of records that are stored across a Spark cluster in a distributed manner. RDDs are immutable, which is to say, they cannot be altered once created. RDDs that are stored across nodes can be accessed in parallel, and hence support parallel operations natively.

The user does not need to write separate code to get the benefits of parallelization but can get the benefits of actions and transformations of data simply by running specific commands that are native to the Spark platform. Because RDDs can be also stored in memory, as an additional benefit, the parallel operations can act on the data directly in memory without incurring expensive I/O access penalties.