Monitoring the Impact of P2P Users on a Broadband Operator's Network over Time

Since their emergence peer-to-peer (P2P) applications have been generating a considerable fraction of the over all transferred bandwidth in broadband networks. Residential broadband service has been moving from one geared towards technology enthusiasts and early adopters to a commodity for a large fraction of households. Thus, the question whether P2P is still the dominant application in terms of bandwidth usage becomes highly relevant for broadband operators. In this work we present an adaption to a previously published method for classifying broadband users into a P2P- and a non-P2P group based on the amount of communication partners ("peers") they have in a dedicated timeframe. Based on this classification, we derive their impact on network characteristics like the number of active users and their aggregate bandwidth. Privacy is assured by anonymization of the data and by not taking into account the packet payloads. We apply our method to real operational data collected 2007 and 2010, respectively, from a major German DSL provider's access link which transported all traffic each user generates and receives. In 2010 the fraction of P2P users clearly decreased compared to previous years. Nevertheless we find that P2P users are still large contributors to the total amount of traffic seen especially in upstream direction. However in 2010 the impact from P2P on the bandwidth peaks in the busy hours has clearly decreased while other applications have a growing impact, leading to an increased bandwidth usage per subscriber in the peak hours. Further analysis also reveals that the P2P users' traffic still does not exhibit strong locality. We compare our findings to those available in the literature and propose areas for future work on network monitoring, P2P applications, and network design.