Live Video Streaming Service over Peer to Peer Network: Design, Implementation and Experience
Yun Tang,Lifeng Sun,Meng Zhang,Shiqiang Yang,Yuzhuo Zhong
2006-01-01
Abstract:Providing live video streaming service over peer to peer network to a large population of end users remains challenging and interesting, due in large part to the scalability of "self-growing" community of peers, high demanding bandwidth over fluctuant and heterogeneous underlay network, system stability to high churn rate as well as stringent deadline for continuous playback. Although there have been many research efforts advocate various algorithms with respect to application layer multicast, overlay or peer to peer network, few existing works indeed study the impact of the congestion on data transmission paths in such networks which is the key to the performance of streaming applications. Besides, it is more important to investigate practical issues in this arena rather than to follow prescribed theoretical assumptions. Thus in this paper we first consider the spanning tree in multicast as the representative transmission mode and then develop a statistical link model to quantitatively analyze the impact of congestion on the system performance. Since the tree-based structure is vulnerable to network congestion and peer dynamics, while link dependence favors system stability, in the design and implementation of a practical live video streaming system, we adopt a gossip-based overlay construction protocol to accommodate topology change and a composite but efficient streaming mechanism to stream video contents with lower delay than simplex data-driven scheme. The experimental results over PlanetLab indicate that this architecture offers continuous playback, demanding bandwidth and low latency. More interestingly, with the service to totally more than 500,000 users and maximum 15,239 concurrent users provided by only one common streaming server in Feb. 2005, we further exhibit insightful statistical results to reveal system properties, and thus demonstrating that it could scale to reliably support a large and highly dynamic population of end users in the Internet..