An optimized load balance based on data popularity on HBASE

Load balancing is an important technique for improving distributed system performance. It results in a better utilization of server resources and quick response time for user requests. In distributed database, it is crucial to take data popularity into account. It is known that servers storing popular data usually become hotspots, so using forecasting methods to predict the popular data in advance will have great impact on performance. For HBase, several methods are proposed for balancing the load but still some improvement in terms of efficiency is required. In this paper, we present an efficient load balancing algorithm which alleviates hotspots by considering data popularity using secondary exponential smoothing method. In the algorithm we construct a region move evaluation function which takes data locality into consideration for further improvement of the throughput in HBase. The experimental result illustrates that our optimized algorithm runs faster and better, and gains about 3% throughput improvement as the cluster is much more balanced.