Determining the Optimal Amount of Computation Pushdown to Minimize Runtime for a Cloud Database

Abstract

Many cloud databases separate their compute from their storage resources. This design introduces a network bottleneck during query execution that can be mitigated through caching and computation pushdown. Depending on the environmental settings and the specific query, the amount of computation pushdown needed to achieve the optimal runtime may vary. This work presents a runtime prediction model that determines the amount of computation pushdown that results in the fastest runtime and analyzes a real-world implementation of this model on the FlexPushdownDB system running in AWS.