Web缓存层次模型与代价分析

Web Caching Hierarchical Model and Cost Analysis

Web缓存分层结构在避免单点失效、提高缓存性能方面具有重要作用.论文研究了Web缓存层次模型,提出请求分发的三种模式,并利用代价函数分析探讨了缓存模型性能.根据Web访问共同特征,实验采用数学建模方法生成模拟日志,模拟不同层采用不同替换算法(LRU、LFU、GDS)时的缓存性能.结果表明,模拟日志的高频区、低频区流行度访问特征分别服从齐普夫第一定律、第二定律,具有真实日志的特性,能够模拟用户请求评价Web缓存层次模型性能;当低层代理缓存采用LFU或LRU替换算法,高层代理缓存采用GDS替换算法时,两层缓存模型在命中率、字节命中率方面有较好的性能表现.

Web proxy caching hierarchy plays a significant role in avoiding single point of fault and improving caching performance.Focusing on the modeling of Web proxy hierarchy caching,this paper presents three different models of how to distribute the requests,and it applies cost function to analyze Web caching performance.Based on several Web common workload characteristics,the experiments make use of the synthetic workloads,which are generated by the method of mathematic modeling,and evaluate the Web cache performance of different combinations of replacement policy（LRU,LFU,GDS） across different levels within Web hierarchy caching model.The experimental results demonstrate that the synthetic workloads,which have the similar performance characteristic with the real traces,follow Zipf＇s 1nd law and Zipf＇s 2nd law and evaluate Web hierarchy caching performance by synthesizing Web object access.When the lower level uses the LFU or LRU replacement policy and the upper level uses the GDS,the two-level hierarchical proxy model achieves higher caching performance in terms of hit ratio and byte hit ratio.