In any distributed processing engine, sending data across a network is the most expensive, as it involves serialization, disk I/O and Network I/O. Shuffling is the step where data needs to be sent across the network. Data shuffle is required when Wide operations and Join operations occur in a Spark job.
