链路层分类包的网络流量自相似性研究  

Study of self-similarity characteristic of network traffic based on classified packets of data link layer

作　　者：聂得欣[1] 袁小坊[1,2] 王东[1] 谢高岗[2] 

机构地区：[1]湖南大学计算机与通信学院,长沙410082 [2]中国科学院计算技术研究所下一代互联网研究中心,北京100080

出　　处：《计算机工程与应用》2009年第5期129-131,146,共4页Computer Engineering and Applications

基　　金：国家自然科学基金网络与信息安全重大专项(No.90604015);国家重点基础研究发展规划(973)(No.2007CB310702);湖南省自然科学基金(No.06JJ4078)~~

摘　　要：提出一种从链路层分类包流量的角度研究网络流量自相似性的方法。使用优化的R/S(rescaled range)法计算Hurst指数,发现分类包流量和总流量一样呈自相似性,并用ON/OFF网络流量模型解释分类包流量自相似的物理原因;并使用主成分分析法研究分类包流量对总流量自相似性的影响,得出大于512B的分类包(大象包)是影响总流量自相似性的主要原因。实验表明该方法是快速有效的。A novel method of studying self-similarity of network traffic is presented.By measuring online the network traffic of classified packets of a backbone link layer in OC-48 POS in a metro area network in long-term,the Hurst exponents of elassi fied packets are been estimated by the method of R/S（rescaled range）.The network traffic of the classified packets is self-similarity and the reason is explained by the ON/OFF traffic model.Then the influence of the traffic of classified packets on the self-similarity of the total traffic is researched by the method of PCA （Primary Component Analysis） and find that the self-similarity characteristic of the total traffic is mainly caused by the classified packets whose lengths are under 512 byte.Finally,the result of the experiment shows that the method is effective.

关 键 词：网络流量 自相似性 HURST指数 R/S法 主成分分析法 

分 类 号：TP393.06[自动化与计算机技术—计算机应用技术]