Towards an energy-proportional storage system using a cluster of wimpy nodes

Abstract

Previous DB research clearly concluded that the most energy-efficient configuration of a single-server DBMS is typically the highest performing one. This observation is certainly true if we focus in isolation on specific applications where the DBMS can steadily run in the peak-performance range. Because we noticed that typical DBMS activity levels-or its average system utilization-are much lower and that 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. We will show that energy proportionality of a storage system can be approximated using a cluster of nodes, built of commodity hardware. To simulate data-intensive workloads, synthetic benchmarks submit read/write requests against a distributed DBMS (WattDB) and, in turn, its HDD- and SSD-based storage system, where time and energy use are captured by specific monitoring and measurement devices. The cluster dynamically adjusts its configuration such that energy consumption and performance are tuned to fit the current workload. For each benchmark setting, an optimal number of nodes is processing the queries in the most energy-efficient way, which does not necessarily correspond to the best performing configuration. The chosen workload is rather simple and primarily serves the purpose to deliver a proof of existence that energy proportionality can be approximated for certain kinds of query processing and, especially, for storage systems.