【摘要】 针对Web集群服务器请求内容的分布特点,提出了一种基于关联内容的请求分发(ACARD)算法.该算法首先以缓存大小为限制,采用FP tree方法挖掘用户的Web使用模式,并把它们所对应的文档映射到不同的后端服务器,然后由前端服务器根据请求的内容匹配相应的模式并进行转发.这种文档的自适应分布提高了集群系统的磁盘利用率和缓存命中率.实验表明,基于内核态的请求分发算法的性能远优于基于用户态的算法,而且在内核态下,ACARD相对加权最小连接算法的系统平均吞吐量可提高20 4%.更多还原

【Abstract】 An efficient associated-content-aware request distribution (ACARD) algorithm was proposed through analyzing the traits of requests distribution in Web cluster server. The algorithm mines periodically the use patterns of Web server with FP-tree method under the constraint of cache size, and then allocates documents of patterns in different back-end servers and forwards client requests to related servers according to their contents, so that the cluster system can achieve high disk usage and cache hit ratio. Experimental results show that the performance of the distributing algorithms in kernel level is much better than that in user level and ACARD can increase the average throughput of the system by 20.4% compared to weighted least-connection.更多还原