Empirical analysis and modeling of peer-to-peer traffic flows

This study presents a detailed flow-level empirical analysis of Peer-to-peer traffic. It is based on 24 hours long traffic traces that were collected on a backbone link in academic environment and in ADSL access. The flows were identified by 5-tuples and were categorized to three main categories: Peer-to-Peer, World-Wide-Web and TCP-big. Four main flow parameters were statistically analyzed: flow interarrival times, size of flows measured in number of packets, size of flows measured in number of bytes, and flow duration. Using the distribution fitting techniques we show that flow interarrival time can be successfully modeled by Weibull distribution, flow size by Pareto distribution and flow duration by log-normal distribution. The key distribution parameters are identified. They indicate a strong resemblance of Peer-to-Peer and TCP-big categories and significantly deviate from the parameters of the World-Wide-Web category. The results can be used in the simulations and other of further studies that involves Peer-to-Peer traffic.