Database servers consume a tremendous amount of energy to operate, impacting not just the bottom line in terms of the cost of running a database server/service, but also limiting deployment of large database appliances in situations where the amount of power available is a constraint. The trend towards decreasing performance per watt in processors, and the definitive shift towards database servers/appliances with large (power hungry) main memory modules, is likely to increase the energy consumption of database servers in the future. Today, database systems do not optimize, or even consider, the energy consumption when planning and executing queries. Consequently, they miss many opportunities for energy efficiency.
Our work is based on recognizing two trends. First, in many database deployments, especially in cloud environments, there are implicit or explicit performance targets that are set for database workloads. Thus, database systems have the freedom to make decisions that potentially save energy provided they still meet the performance targets. Second, modern hardware is increasingly being built with mechanisms that allow software to explicitly control the performance and energy states of the hardware. We have developed a framework that explicitly incorporates the expected amount of energy consumed to the query optimization and execution framework of a typical relational database engine. As a proof-of-concept, we have implemented this framework in a stand-alone instance of SQL Server, and show real reduction in end-to-end energy consumption of queries [LKP11]. On going work in this area is now expanding this work to cluster and cloud environments running database services.
- Willis Lang
- Jignesh Patel
- Nikhil Teletia
Willis Lang, Stavros Harizopoulos, Jignesh M. Patel, Mehul A. Shah, Dimitris Tsirogiannis: Towards Energy Efficient Database Cluster Design, PVLDB 5(11): 1684-1695 (2012), Proceedings of the 2012 VLDB Conference, Istanbul, Turkey, August 2012.
Willis Lang, Ramakrishnan Kandhan, Jignesh M. Patel: Rethinking Query Processing for Energy Efficiency: Slowing Down to Win the Race. IEEE Data Eng. Bull. 34(1): 12-23 (2011).
Willis Lang, Jignesh M. Patel, Srinath Shankar: Wimpy node clusters: what about non-wimpy workloads?, DaMoN 2010: 47-55.
Willis Lang, Jignesh M. Patel: Energy Management for MapReduce Clusters. PVLDB 3(1): 129-139 (2010)
Willis Lang, Jignesh M. Patel: Towards Eco-friendly Database Management Systems. CIDR 2009.
Willis Lang, Jignesh M. Patel, Jeffrey F. Naughton: On energy management, load balancing and replication. SIGMOD Record 38(4): 35-42 (2009)