What is the primary effect of broadcasting a small DataFrame when joining with a large DataFrame in Spark?

• A. Increases shuffle
• B. Increases memory usage and reduces network shuffles
• C. No effect on performance
• D. Forces a repartition of data

Answer: (B)

Broadcasting a small DataFrame means sending its data to every executor so the join can be performed locally on each partition of the large DataFrame. This enables a broadcast hash join, allowing each worker to join its portion of the big DataFrame with the small one without shuffling the large dataset across the cluster. The primary effect is increased memory usage on each executor to hold the broadcasted data, while network shuffles are reduced because the join happens locally rather than repartitioning the large DataFrame. If the small DataFrame truly fits in memory, this often speeds up the join; if it’s too large or memory constrained, it can lead to memory pressure.