基于无监督学习的P2P流量识别  

P2P Traffic Identification Based on Unsupervised Learning Approach

作　　者：贾波[1] 邹园萍[1] 

机构地区：[1]浙江工商大学信息与电子工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2011年第4期148-151,共4页Journal of Hangzhou Dianzi University：Natural Sciences

摘　　要：该文以应用层流量分类为重点,分析了流量分类中的一些关键问题;为了适应现代网络管理,实现对应用层流量的实时监控,针对现有方法存在的问题,提出了一种分层次的应用层流量识别方法。为了具有对新应用的发现能力,结合基于深层包检测和基于数据流特征的方法,并同时使用基于无监督学习技术,发现P2P等新的网络应用。通过对校园网流量的跟踪,证明了该方法的有效性。结果表明,识别流行的P2P应用具有较高的准确性。This paper put emphasis on application-level traffic classification, analyzed a few critical issues ac- companying with traffic iidentification first then a number of shortfalls of the traffic identifying approaches be- ing used or researched are analyzed. Finally ＇an approach of network traffic identification is proposed, witch combine deeply packet inspection and based on features .of data flow with unsupervised learning to identify cut- rent sophisticated traffiei data generated from various emerging network applications. It can implement real-time traffic monitoring of traffic of the appliehtionqevel and the ability to discover new applications at the same time. We demonstratedtthe effectiveness of our method on campus traffic tracking. The result indicates that this approach can identify the popular P2P applications with higher accuracy.

关 键 词：流量识别 数据挖掘 无监督学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]