可区分服务的反馈式Web应用负载均衡策略

On reactive load balancing strategy with service differentiation for Web applications

为研究Internet环境下Web应用面向区分服务的性能优化策略,从多个方面提出了动态反馈式负载均衡的改进方案,包括基于K-means算法的预处理、双周期式负载反馈、基于反馈的负载距离计算和引入随机概率的节点选择等,在资源层面上对用户请求和服务节点进行区分并把握它们之间的关联,实现更有针对性的调度。实验分析结果表明,与传统的策略进行对比,改进策略能够降低请求的平均响应时间和请求失效数,在保障负载均衡的服务质量上取得效果。

To research the performance of optimization strategy for Web applications with service dif- ferentiation requirements on Internet, multiple improvement methods for reactive dynamically load balancing strategy are presented, including pretreatment based on K-means algorithm, double peri- odic load feedback, load distance calculation based on feedback, and node selection with random probability. In order to implement targeted load scheduling, the proposed strategy distinguishes the types of user request and server nodes at resource layer and learns the relations between user request and backend nodes reasonably. Experiment results show that, the improved strategy can help to shorten the average response time and to reduce the numbers of failed requests, achieving quality guarantee of load balancing.