Energy efficiency has been a critical issue in database clusters. In this paper, we present an energy-proportional database cluster and propose an energy-proportional query processing approach to reduce the energy consumed by database clusters while keeping high time performance. Particularly, we introduce a
query stream buffer
on top of a database cluster and propose an
unbalanced load allocation
algorithm to distribute workloads among the cluster so as to realize better energy proportionality. Further, we present an adaptive algorithm to turn on/off nodes according to workload changes. With this mechanism, we can reduce energy consumption while keeping high time performance for query processing on database clusters. We build a prototype database cluster and use the TPC-H benchmark to compare our proposal with three baseline methods, where different query patterns are used. The results suggest the superiority of our proposal in energy savings and time performance.