FlexPushDownDB paper appeared at VLDB. It investigates the tradeoff between caching data at the query execution server vs. pushing computation to storage in analytical query workloads.