面向流量识别系统的聚类算法的比较与分析  

Compare and Analysis of Uastering Algorithms Oriented Tvaffic Identification System

作　　者：苏欣[1] 杨建华[2] 张大方[1] 谢高岗[2] 

机构地区：[1]湖南大学软件学院,湖南长沙410082 [2]中国科学院计算技术研究所,北京100080

出　　处：《计算技术与自动化》2008年第3期1-6,共6页Computing Technology and Automation

基　　金：国家自然科学基金资助(60673155;90718008)

摘　　要：考虑到基于端口的识别方法准确性比较低,而基于有效负载的方法的开销太大,促使利用应用连接到网络时的特征流的特点来识别流量。引用几种常见的聚类算法,提出基于聚类算法的识别方法,该方法包括两个阶段:通过离线学习阶段学习到流量的特征;在线识别阶段把学习到的特征用于在线的流量识别。通过实验表明该方法对流量识别,尤其是P2P流量最高可达90%以上的识别率。The diminished effectiveness of port - based identification and the overheads of payload- based approaches motivate us to identify traffic by exploiting distinctive flow characteristics of applications when they communicate on a network. In this paper, we import several clustering algorithms, and propose an identification method based on clustering algorithm. This method includes two phases： learn the character of traffic in offline phase, and use the character in realtime traffic identification. Experiment shows that the accuracy of traffic identification, especial for P2P traffic, could achieve 90 %. or more.

关 键 词：聚类 流量识别 学习 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]