Traffic dispersion graph based anomaly detection

Detecting and diagnosing anomalous traffic are important aspects of managing IP networks. In this paper, we propose a novel approach to detect anomalous network traffic based on graph theory concepts such as degree distribution, maximum degree and dK-2 distance. In this approach, we have used the traffic dispersion graphs (TDG) to model network traffic over time. We analyze differences of TDG graphs in time series to detect anomalies and introduce a method to identify attack patterns in anomalous traffic. The approach has been validated by using network traces from POSTECH and CAIDA.