Mining network traffic data

Mining and statistical analysis of network data are often employed to determine traffic loads, analyze patterns of users´ behavior, and predict future network traffic. In this talk, we analyze traffic data collected from three deployed networks: a cellular wireless network used by public safety agancies, a satellite network used by Internet service providers, and the Internet. We describe analysis of network log data collected from E-Comm, a public safety trunked radio network that utilizes circuit-switched cellular wireless technology. We examine statistical distributions and autocorrelation functions of call inter-arrival and call holding times during busy hours. Our findings indicate that traditional Er land models used for voice traffic may not be suitable for evaluating the performance of trunked radio networks. We also describe collection of traffic data, characterization of traffic loads, and distribution of packet sizes in ChinaSat, a hybrid satellite-terrestrial system. We investigate long-range dependence as the traffic patterns very, propose a traffic model for the transmission control protocol (TCP) connections, and use data from billing records to predict future traffic loads. Since discovering network topology is important for analyzing routing protocols and network robustness and resilience, we also examine datasets from the route views and RIPE projects and identify important properties of Internet graphs. Our current traffic measurement project deals with traffic collection from BC. Net, the Vancouver lower mainland gigabit Ethernet network. Collected data will be used to analyze behavior and performance of the border gateway protocol (BGP).