Characterizing Locality-aware P2P Streaming

Peer-to-peer (P2P) live streaming systems have been increasingly popular and successful in today's Internet, which provide large collections of video channels to millions of users at low server costs. The large volumes of P2P streaming traffic are exceeding those incurred by BitTorrent-like file sharing applications, threatening huge traffic relay cost to the Internet service providers (ISPs). There have recently emerged proposals advocating locality-aware P2P streaming protocol design, which aim to constrain streaming traffic within ISP boundaries and to alleviate traffic relay cost to the Internet Service Providers (ISPs). Nevertheless, there is a lack of in-depth understanding on the impact of such a locality-aware design on P2P streaming performance. Taking an analytical approach, we model the relation between streaming performance and traffic locality in P2P live streaming systems, in order to acquire useful insights for designing high-performance locality-aware real-world systems. We use end-to-end streaming delays as the performance metric for live streaming, and the number of copies of the live streams imported into an ISP to evaluate the volume of inter-ISP traffic. Considering multiple ISPs at different bandwidth levels, we characterize the generic relationship between the volume of inter-ISP traffic and the streaming performance; we then analyze the traffic volume when the best streaming performance is achieved and the streaming performance when minimum inter-ISP traffic is incurred. Our models and analyses provide intriguing insights on the design of effective locality-aware peer selection protocols and server deployment strategies across multiple ISPs. We also evaluate our models and theoretical results with large-scale simulations under realistic settings.