Rethinking Query Processing for Energy Efficiency: Slowing Down to Win the Race.

TLDR

The TPC benchmarks now include an energy‑efficiency metric, reflecting the urgent demand for energy‑efficient database processing and the need for query engines to become energy‑aware to capture savings. This paper designs and evaluates a general framework that optimizes query plans for both performance constraints specified in SLAs and energy consumption. The framework exploits modern hardware’s ability to operate in multiple energy and performance states, uses an energy consumption model for query operations, and incorporates a model of a commercial DBMS. Experiments show the optimizer can meet SLA performance while lowering energy use, achieving significant system‑wide savings and pointing to larger opportunities with upcoming energy‑aware technologies.

Abstract

The biggest change in the TPC benchmarks in over two decades is now well underway – namely the addition of an energy efficiency metric along with traditional performance metrics. This change is fueled by the growing, real, and urgent demand for energy-efficient database processing. Database query processing engines must now consider becoming energy-aware, else they risk missing many opportunities for significant energy savings. While other recent work has focused on solely optimizing for energy efficiency, we recognize that such methods are only practical if they also consider performance requirements specified in SLAs. The focus of this paper is on the design and evaluation of a general framework for query optimization that considers both performance constraints and energy consumption as first-class optimization criteria. Our method recognizes and exploits the evolution of modern computing hardware that allows hardware components to operate in different energy and performance states. Our optimization framework considers these states and uses an energy consumption model for database query operations. We have also built a model for an actual commercial DBMS. Using our model the query optimizer can pick query plans that meet traditional performance goals (e.g., specified by SLAs), but result in lower energy consumption. Our experimental evaluations show that our system-wide energy savings can be significant and point toward greater opportunities with upcoming energy-aware technologies on the horizon.