基于UDP流量的P2P流媒体流量识别算法研究  

Research on P2P streaming media identification based on UDP

作　　者：董仕[1,2,3,4] 王岗[2,3,4] 

机构地区：[1]周口师范学院计算机科学与技术学院,河南周口466001 [2]东南大学计算机科学与工程学院,江苏南京210092 [3]江苏省计算机网络技术重点实验室,江苏南京210092 [4]国家教育部计算机网络和信息集成重点实验室,江苏南京210092

出　　处：《通信学报》2012年第12期25-34,共10页Journal on Communications

基　　金：国家重点基础研究发展计划("973"计划)基金资助项目(2009CB320505);国家科技支撑计划基金资助项目(2008BAH37B04)~~

摘　　要：以几款主流的P2P流媒体网络电视作为研究对象,深入分析了其产生的流量在端口使用方面的特点和报文长度分布上的差异。通过对这些特征的总结和提取,获得了基于端口特性"在一次交互过程中,特定主机的特定端口唯一确定一种应用"等结论。在此基础上提出了一种基于带有扩展属性的流记录准确识别P2P应用UDP流量的EXID算法。通过对CERNET江苏省边界10G主干信道上采集的Trace数据中5种P2P流媒体应用进行识别,并与机器学习流量识别算法进行比较,其结果表明提出的方案具有很高的查准率和查全率,时间效率高,且不易受样本比重的影响。Several popular P2P networks TV were studied, and their differences in port usage and packet size distribution were analyzed thoroughly. By observing the above characteristics, the conclusions that network TV application employs only one port to generate most of UDP traffic in one communication period were summarized, and the UDP packet sizes in various network TV differ significantly. Thus, a method that can identify P2P application＇s UDP traffic accurately and effectively was proposed based on extended flow records. Through identifying and verifying the five P2P streaming application traffic which was called Trace data collected from the backbone channel of CERNET （China education and research network） border in Jiangsu Province, and traffic identification results compared with machine learning algorithins show that the proposed method has a high precision rate and recall rate, high time efficient, and not susceptible to the impact of the proportion of the sample.

关 键 词：UDP P2P 流量识别 扩展的流记录 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]