基于灰预测的Web Cache集群热点对象处理策略
Grey Prediction Based Hot Spot Relief Strategy in Web Cache Cluster
-
摘要: 热点对象问题是Web cache集群负载均衡研究中的一个重要问题,热点对象的存在容易引起访问歪斜(access skew)和负载不均衡. 在分析Web 访问中热点对象特征的基础上,构建了基于灰预测的热点对象访问频率预测模型. 基于该模型,提出一种新的热点对象处理策略:预测对象的访问频率,配合阈值判断对象是否为热点对象,将热点对象推送到集群中所有节点,实现热点对象的冗余存储,从而达到负载均衡. 实验结果表明,与其他热点对象处理策略相比,基于灰预测的热点对象处理策略可以明显提升Web cache集群性能.Abstract: Hot spot is an important problem in load balancing research of Web cache cluster. The existence of hot spot could easily induce access skew and load imbalance. After analyzing features of hot spot, a model to predict hot spot access frequency is built based on grey prediction. The proposed procedure of new hot spot relief strategy can be described as follows: set a threshold, distinguish object, whose predicted access frequency exceeds the threshold, as hot spot, push hot spot into all nodes, implement redundant store for hot spot, and finally achieve the load balance in Web cache cluster. Compared with other strategies, grey prediction based hot spot relief strategy heightens the performance of Web cache cluster.
下载: