We know some memory of spark are used for computing (e.g., shuffle buffer) and some are used for caching RDD for future use. Is there any existing workload which utilize both of them? I want to do some performance study by adjusting the ratio between them.