P2P应用流量的高效分类方法研究  

RESEARCH ON EFFICIENT CLASSIFICATION METHOD OF P2P APPLICATION TRAFFIC

作　　者：陈金富[1,2] 赵慧[3] 常鹏[1] 张永铮[1] 

机构地区：[1]中国科学院信息工程研究所,北京100093 [2]中国科学院大学,北京100049 [3]国家计算机网络应急技术处理协调中心,北京100029

出　　处：《计算机应用与软件》2017年第4期110-117,156,共9页Computer Applications and Software

基　　金：国家自然科学基金项目(61572496);国家高技术研究发展计划基金项目(2013AA014703)

摘　　要：随着互联网应用的广泛使用,网络应用已经呈现出很多类别,尤其是P2P应用流量的暴增。传统的流量分类和应用识别方法已经达不到稳定可观的应用识别率。为了提高P2P应用流量分类准确率和稳定性,科学管理规划网络,提出WMFA(滑动窗口多流关联)分类算法,使用P2P应用流量统计特征,通过降低流统计特征维数,以及减少计算每个流中包的数量,利用C4.5决策树算法对P2P主流应用进行一次分类,采用WMFA算法进行误识别流的挖掘,再进行多流关联进行二次识别,从而提高P2P应用流量分类准确率。实验表明,在降低流特征维数以及减少每个流数据包的前提下,面向国内主流P2P应用WMFA算法对P2P应用在线识别的分类正确率达到96%以上,在准确率上比现有方法平均提高3%。Network application has been showing a lot of categories because of the widespread use of Internet applications, especially P2P applications traffic growth. Traditional traffic classification and application identification methods can’t considerable reach stable application classification precision. In order to improve classification accuracy and stability of P2P application traffic and manage network scientifically, this paper proposes WMFA （sliding window multi-flow association） algorithm. Using P2P application traffic statistics feature by reducing the flow statistics feature dimension and reducing the number of packets in each network traffic flow, C4.5 algorithm is used to classify P2P applications. We use WMFA algorithm to mine misrecognized flows, and carry on the multi-flow association on the second recognition to improve the P2P application traffic classification accuracy. Experimental results show that with a decrease in P2P traffic flow characteristics dimension and reducing the number of each flow data packets, WMFA algorithm average classification precision for the domestic mainstream P2P application is more than 96%, the average rate of accuracy than the existing method of 3%.

关 键 词：P2P流量分类 应用识别 WMFA算法 多流关联 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]