In the event your domain knowledge can help you make decisions about how spark load-balances or stripes data across worker nodes. Persistence “you should store this data in faster/slower memory” MEMORY_ONLY, MEMORY_ONLY_SER, MEMORY_AND_DISK, MEMORY_AND_DISK_SER, DISK_ONLY rdd.persist(StorageLevel.MEMORY_AND_DISK) # ... do work ... rdd.unpersist() Parallel Programming