基于隐马尔科夫模型的P2P流识别技术

Hidden Markov model based P2P flow identification technique

为了实时、准确地识别多种P2P应用流,提出了基于隐马尔科夫模型(HMM,hidden Markov model)的P2P流识别技术。该技术利用分组大小、到达时间间隔和到达顺序等特征构建流识别模型,采用离散型随机变量刻画HMM状态特征;提出了能同时识别多种P2P应用流的架构HMM-FIA,设计了HMM的状态个数选择算法。在校园网中架设可控实验环境,使用HMM-FIA识别多种P2P流,并与已有识别方法进行比较,结果表明采用离散型随机变量能降低模型建立时间,提高识别未知流的实时性和准确性;HMM-FIA能同时识别多种P2P协议产生的分组流,并能较好地适应网络环境变化。

To identify various P2P flows accurately in real-time,a hidden Markov model（HMM） based P2P flow identification technique was proposed.This approach made use of packet size,inter-arrival time and arrival order to construct flow identification model,in which discrete random variable was used to depict the characteristics of HMM state.A framework called HMM-FIA was proposed,which could identify various P2P flows simultaneously.Meanwhile,the algorithm for selecting the number of HMM state was designed.In a controllable experimental circumstance in the campus network,HMM-FIA was utilized to identify P2P flows and was compared with other identification methods.The results show that discrete random variable can decrease the model constructing time and improve the time-cost and accuracy in identifying unknown flows,HMM-FIA can correctly identify the packet flows produced by various P2P protocols and it can be adaptive to different network circumstance.