Quantitative Analysis of Strategies for Streaming Media Distribution

Distribution of streaming media content, including live news, music and videos, is becoming increasingly popular in today’s Internet. Traditional client/server architectures are inefficient for distributing streaming media objects because of the high demands for system resources, especially server and network bandwidth, which severely limit the total number of simultaneous users the system can support. One proposal for improving the scalability of media distribution systems is the use of P2P overlay networks. Although a number of previous work has evaluated different aspects of P2P systems, mainly through simulation, there is a lack of a thorough quantitative analysis of the requirements for server and network resources (i.e., CPU, server and network bandwidth) in actual P2P systems, compared to traditional client/server systems. This work aims at filling this gap by providing experimental results that quantify the savings in server and network resources if a P2P approach is used for distributing live streaming media instead of the traditional client/server approach. Towards this goal, we build an experimental testbed, in a controlled environment, to evaluate actual systems with varying number of clients during periods when the distribution tree is static. A key component of this experimental testbed is a new efficient and scalable application called streaming servent, which can act both as a client and a server, forwarding packets to other clients. We also use simple analytical formulas to evaluate the scalability of our servent application. The experimental results quantify the intuitive better scalability of the P2P architecture. As an example, the total server bandwidth decreases from 15 Mbits/s to 9 Mbits/s (a 40% reduction) if a P2P architecture is used instead of a client/server architecture for live delivery of a given file to 24 clients.