Network flow classification based on the rhythm of packets

Abstract

Accurate traffic classification is a necessary means of network management, QOS, monitoring and so on. We find that each protocol's flows have their own packet-level rhythm on the statistical characteristics. In this paper we present a Bayesian network classification mechanism based on the flows' packet-level rhythm. However, the flows rhythm is always too scattered to bring into play its ability well in the Bayesian network, so we employ an Equal-width discretization method to centralize the rhythm and discretize the packet size and interval-time to some different space. Then we applied our classification model to the different discretization data set of HTTP, EDONKEY, BITTORRENT, FTP and AIM. Experiment results show that our approach can achieve better precision and recall rate for these applications. © 2011 Springer-Verlag.