In Apache Spark, which caching level should you use to ensure a DataFrame is stored exclusively in memory for optimal performance?