基于NetFlow流记录的高速应用流量分类方法

针对目前应用流量分类算法效率不高的现状,提出一种以NetFlow统计的IP流记录信息作为输入的高速应用流量分类(FATC,fast application-level traffic classification)算法。该算法采用基于简单相关系数的测度选择算法衡量测度变量间的相关关系,删除对分类无用或相互冗余的测度,而后使用基于Bayes判别法的分类算法将网络流量分至误判损失最小的应用类别中。理论分析及实验表明,FATC算法在具有超过95%的分类准确率基础上,极大降低了当前应用流量分类方法在训练和分类过程的时空复杂度,满足实时准确分类当前10Gbit/s主干信道网络流量的需求。

Fast application-level traffic classification using NetFlow records

In order to improve the performance and reduce the resources usage of application-level traffic classification,a novel fast application-level traffic classification(FATC) algorithm using IP flow record from NetFlow as input was pre-sented.FATC adopted metric selection algorithm based on correlation coefficient to measure the correlation among flow metric variables,and deleted the irrelevant or redundant metrics,then used Bayes discrimination to classify network traf-fic to the application category that of smallest misjudge loss.The theoretical analysis and experimental results show that,with more than 95% accuracy,the FATC algorithm greatly reduces the time and space complexity of current applica-tion-level traffic classification algorithms during the training and classification processes,and can work efficiently on 10Gbit/s backbone network in real time.