What options are there in speeding up the compute-plan step? Is it possible to provide a way to cache the compute plan for common queries?