Abstract
Emerging Peer-to-Peer (P2P) technologies have in the past few years enabled various types of content, such as file sharing and live video streaming, to be efficiently distributed over the Internet. However, the uncontrolled behaviour of P2P applications in consuming Internet bandwidth leads to the situation where P2P flows account for some 50%-70% of the overall Internet traffic. On the one hand, traditional Traffic Engineering (TE) techniques are deployed by Internet Service Providers (ISPs) to map traffic onto the network for achieving overall network performance optimisation, without distinguishing between standard traffic and P2P traffic. On the other hand, recent proposals have been made on the cooperation between P2P applications and the underlying ISP to not only make best use of network resources but also improve application performance, such as Application Layer Traffic Optimisation (ALTO). In this thesis we investigate how to design future intelligent Internet P2P traffic management paradigms in both a non-cooperative way (i e. an ISP-centric solution) and in a cooperative way (e.g. ALTO). Some key contributions are summarised below: 1. We investigate whether ALTO can synergistically coexist with an application-agnostic TE (AATE). We evaluate whether ALTO is an acceptable alternative for optimizing network resources, and we consider the behaviour of distinct P2P overlays (non-, semi- and fully-cooperative) coexisting with AATE, and the impact of various traffic scenarios on both application and network sides. 2. We show how to improve the performance of non-P2P services while accommodating as much P2P traffic as possible (ISP-centric solution). Since conventional Internet services (e.g. web browsing) are significantly impacted by P2P traffic and the market reputation of the ISP may be impacted if P2P traffic is absolutely blocked, we propose a dynamic P2P traffic limiting policy for ISP networks to achieve a better trade-off between ISPs and P2P systems. 3. We demonstrate significant improvement in bandwidth resource utilisation when using the cooperative approach. Due to lack of network condition information and unintelligent peer selection, current approaches deliver significant amounts of unnecessary P2P traffic. We propose an adaptive peer selection scheme that is aware of dynamic network conditions and a localised P2P traffic exchange approach to reduce P2P traffic and achieve network load balancing.