Energy and performance - can a wimpy-node cluster challenge a brawny server

Abstract

Traditional DBMS servers are often over-provisioned for most of their daily workloads and, because they do not provide energy proportionality, waste more energy than necessary. A cluster of wimpy servers, where the number of nodes can dynamically adjust to the current workload, might offer better energy characteristics for these workloads. Yet, clusters suffer from friction losses and cannot quickly adapt to the workload, whereas a single server delivers maximum performance instantaneously. Designed for a cluster of nodes, our WattDB system primarily aims at energy proportionality for a wide range of DB applications. In this paper, we check this system under OLTP and OLAP workloads against a single-server DBMS in terms of throughput/response time and energy efficiency. To test the system's ability to adjust to changing workloads, we execute several benchmark at differing system activity levels. To quantify possible energy saving and its conceivable drawback on query runtime, we evaluate our WattDB implementation-to obtain maximum accuracy possible-on a cluster of wimpy nodes as well as on a single, brawny server and compare the results w.r.t. performance and energy consumption. Our findings confirm that-especially for OLAP workloads-energy can be saved without sacrificing too much performance.