Towards Eco-friendly Database Management Systems

TL;DR

This paper introduces two techniques, PVC and QED, that enable database management systems to reduce energy consumption during query processing by trading off performance, demonstrated on commercial systems with significant energy savings.

Contribution

The paper presents two novel methods, PVC and QED, for directly managing energy consumption in DBMSs, addressing an overlooked aspect of query processing performance.

Findings

PVC reduces energy by up to 49% with minimal response time increase

QED achieves 54% energy savings with a 43% response time increase

Experimental validation on MySQL and commercial DBMSs confirms effectiveness

Abstract

Database management systems (DBMSs) have largely ignored the task of managing the energy consumed during query processing. Both economical and environmental factors now require that DBMSs pay close attention to energy consumption. In this paper we approach this issue by considering energy consumption as a first-class performance goal for query processing in a DBMS. We present two concrete techniques that can be used by a DBMS to directly manage the energy consumption. Both techniques trade energy consumption for performance. The first technique, called PVC, leverages the ability of modern processors to execute at lower processor voltage and frequency. The second technique, called QED, uses query aggregation to leverage common components of queries in a workload. Using experiments run on a commercial DBMS and MySQL, we show that PVC can reduce the processor energy consumption by 49% of the original consumption while increasing the response time by only 3%. On MySQL, PVC can reduce energy consumption by 20% with a response time penalty of only 6%. For simple selection queries with no predicate overlap, we show that QED can be used to gracefully trade response time for energy, reducing energy consumption by 54% for a 43% increase in average response time. In this paper we also highlight some research issues in the emerging area of energy-efficient data processing.