Proactive Energy Management in Database Systems

Reducing carbon footprint of datacenters is an unavoidable challenge for sustainable IT infrastructure. As the core software for data processing and management, energy efficiency in database systems has arisen as an important research subject. The tradeoff between power consumption and performance has been extensively studied across various workloads and query optimization techniques have been proposed to improve energy efficiency. Utilization of general-purpose power management methods such as dynamic frequency/voltage scaling and device hibernation in the context of database systems has also been studied. However, many other components that make up database systems still remain largely untouched from the perspective of energy management. Energy-oriented behavioral changes in these components can unlock the potential for further energy management. This paper explores the opportunity that database systems can proactively manage energy consumption during query execution. We particularly focused on the query execution and storage management layers, which largely determine the behavior of the system during query execution. We propose an initial design of the query execution model that proactively manages energy consumption coupled with the energy-aware storage layout. Our prototype using PostgreSQL demonstrated up to 29.6% energy reduction with 7.6% execution time overhead in an empirical evaluation using TPC-H benchmark. We also discuss the future directions of proactive energy management in database systems.