A Novel Method of P2P Hosts Detection Based on Flexible Neural Tree

Abstract

It is estimated that 70 percent or more of broadband bandwidth is consumed by transporting music, games, video, and other content through P2P clients. In order to identify, and manage of P2P traffic, some port, payload and connection features based methods were proposed. But most of them focus on identifying P2P traffic based on rules and most of them can't identify P2P traffic with new features. None of the proposed methods can effectively help controlling and managing the P2P hosts. In this paper, we created a new P2P feature set, which include most of the proposed and new P2P features proposed by us. And a new anormaly detection method based on the feature set and flexible neural tree (FNT) is applied to detect P2P hosts. This approach is the first intelligent P2P hosts detection method. Experimental results show that the new method proposed by this paper has high detection performance and good expansion ability to add new P2P features. It can be easily applied to an on-line environment in our future planed work