WattDB - a Rocky Road to Energy Proportionality

Energy efficiency is becoming more important in database design, i. e., the work delivered by a database server should be accomplished by minimal energy consumption. So far, a substantial number of research papers examined and optimized the energy consumption of database servers or single components. In this way, our first efforts were exclusively focused on the use of flash memory or SSDs in a DBMS context to identify their performance potential for typical DB operations. In particular, we developed tailor-made algorithms to support caching for flash-based databases [3], however with limited success concerning the energy efficiency of the entire database server. A key observation made by Tsirogiannis et al. [5] concerning the energy efficiency of single servers, the best performing configuration is also the most energy-efficient one, because power use is not proportional to system utilization and, for this reason, runtime needed for accomplishing a computing task essentially determines energy consumption. Based on our caching experiments for flash-based databases, we came to the same conclusion [2]. Hence, the server system must be fully utilized to be most energy efficient. However, real-world workloads do not stress servers continuously. Typically, their average utilization ranges between 20 and 50% of peak performance [1]. Therefore, traditional singleserver DBMSs are chronically underutilized and operate below their optimal energy-consumption-per-query ratio. As a result, there is a big optimization opportunity to decrease energy consumption during off-peak times. Because the energy use of single-server systems is far from being energy proportional, we came up with the hypothesis that better energy efficiency may be achieved by a cluster of nodes whose size is dynamically adjusted to the current workload demand. For this reason, we shifted our research focus from inflexible single-server DBMSs to distributed clusters running on lightweight nodes. Although distributed systems impose some performance degradation com-