Research Of Balancing For Hotpot In The Distribute Cluster

Distributed cluster system is widely used to deal with big data.The implement of load balance of cluster system is an guarantee of the stability and efficiency of the cluster system. Although the load balance involves widely, there is little research on load balance of HBase in a hot scene.So there is room for improvement.Based on the analysis of distributed load balancing mechanism and HBase workflow, a load balancing mechanism for hot scene based HBase database platform is designed. The mechanism has three optimization on designing load factor selection, designing a combining and ranking algorithm of hot load balancing and implementing cache moving. Considering software, hardware, load factor research profiles, hot scenes demand, the degree of difficulty to achieve and other factors, the mechanism makes the final choice of the CPU usage, the number of read requests and the response time of reading request as the load factors. The quality of the load factors has much influence on the performance of the load balancing algorithm and the load balancing mechanism. Combining and ranking algorithm of hot load balancing mechanism is designed to compute the balancing value on different level nodes and divide the node into different set by the balancing value. To further optimize the performance of load balancing of hotpot, the cache data belong to the moving node is kept and moved to the other node following the moving node which improves the speed of the node's beginning.Based on the HBase-0.96.0, the load balancing of hotpot has greater advantage in the hotpot scene compared with the HBase load balancing which supplies more throughout by balancing the cluster more quickly.