Peer-assisted content distribution networks: performance gains and server capacity savings

Abstract

Content distribution networks are experiencing tremendous growth, in terms of traffic volume, scope, and diversity, fueled by several technological advances and competing paradigms. Traditional client/server architectures as deployed in the majority of today's commercial networks provide high reliability and superior level of service quality but involve significant investment in network infrastructure. In contrast, peer-to-peer swarming technologies such as BitTorrent have emerged as a popular mechanism for cost-effective and scalable content dissemination, accounting for a huge portion of today's Internet traffic, but stand or fall with the availability of abundant, free bandwidth, rendering performance inherently unreliable. A hybrid approach offers a promising concept to strike the right balance and combine the highly reliable performance of traditional client/server systems with the scalability and cost-effectiveness of peer-to-peer mechanisms. In order to examine the potential performance gains and capacity savings from a hybrid approach, we develop simple mathematical models for peer-assisted content distribution networks. We use these models to investigate the impact of several key parameters, such as churn rate, object size, and upload bandwidth, and to compare various resource management strategies. The results demonstrate that a peer-assisted approach can yield substantial performance gains and capacity savings compared to a pure client/server system, with churn rate and upload bandwidth being critical factors. Compared to a pure peer-to-peer scenario, the hybrid approach can dramatically boost the performance and improve the reliability.