Study Of Method For P2P Traffic Identification Based On Fuzzy ARTMAP

With the fast development of P2P networking technology, especially the expansion of file-sharing and streaming applications,P2P greatly enriched the content of the Internet, but its huge traffic and higher bandwidth posed a serious threat for other network services. However, only by increasing the bandwidth does not alleviate the congestion situation of network.Therefore, effective and accurate identification of P2P traffic becomes the current problem to be settled urgently, and makes great sense for effective management of networks, rational use of the Internet infrastructure and network resources.Firstly,This paper introduces the concept of P2P, and compare the methods for P2P traffic identification P2P traffic at home and abroad.Then, in this paper, statistics and analysis of traffic characteristics of the traditional non-P2P applications and domestic popular P2P applications (Emule, BitTorrent, PPLive, PPstream) from three aspects on packet, data stream and connection are made, and four basic traffic characteristics for traffic identification: positive connection - reverse connection, the data reported average length, the ratio of upstream traffic in total traffic, the ratio of TCP traffic in the total traffic, are slected. The previous three characteristics of P2P traffic and non-P2P traffic have their own independent band; fluctuation intervals of the second and the third characteristic of four kinds of P2P traffic overlap up each other , but between the file sharing application BT and Emule , and between the streaming media transmission application PPLive and PPSream, are different; although the fourth traffic characteristic of PPSream and PPLive is very similar, there is Significant difference among of BT, Emule and PPSream/PPLive.Because of Fuzzy ARTMAP neural network supporting any dimension mapping between the analog input and output space, and its fast and stable characteristics of online identification of learning, we propose a method for P2P traffic identification based on Fuzzy ARTMAP neural network: Firstly, collecting traffic and dividing into training sampleset and testing sample set; secondly, extracting characteristic of the sample sets and normalizing characteristic values into characteristic vectors; thirly, training the classifier based on Fuzzy ARTMAP with training sample; finaly, classifing the test samples with trained tested on identified.Experiment shows that :in the P2P and non-P2P traffic classification using three traffic characteristics of positive connection - reverse connection, average paket length and the ratio of upstream traffic in total traffic, P2P traffic identification accuracy rate is 98%, and anonymous P2P traffic identification accuracy also achieved 96%; when adding the characteristics of the ratio of TCP traffic in the total traffic for multi-application classification, the average accuracy rate is up to 96%, and the average false alarm rate of only is 1.5%. The experimental result shows that the method is scalable, can effectively classify multi-application traffic and identify anonymous P2P traffic.