Performance Analysis and Industrial Practice of Peer-Assisted Content Distribution Network for Large-Scale Live Video Streaming

Abstract

Recently efficient and scalable live video streaming system over the Internet has become a hot topic. In order to improve the system performance metrics, such as startup delay, source-to-end delay, playback continuity and scalability, many previous works developed two successful cases of content distribution network (CDN) and peer-to-peer (P2P) Network for the design of large-scale live video streaming systems, but no single one has yet delivered both the scale and service quality. To combine the advantages of CDN and P2P network has been considered as a feasible orientation for large-scale video stream delivering. In this paper, we propose a peer-assisted content distribution network, i.e. PACDN, which borrows the mesh-based P2P ideas into the traditional CDN to enhance the performance and scalability. The basic features of PACDN include: 1) To meet the real time requirement of live video stream service, i.e. to ensure that the video stream could be continuously and stably delivered from the source to each edge server for offering good QoS to different regions clients, the placement edge servers and source streaming server(s) build a hierarchical multi-tree based and in-hierarchy peer-assisted overlay, which is optimized according to the knowledge of underlying physical topology. This scheme in the design is called "server side peer-assisted". 2) To enhance the system scalability and reduce the deployment cost, clients and edge servers construct a Client/Server based and P2P network assisted overlay with the increasing of viewers, which is called "client side peer-assisted" in this design. We compare the inner performance of PACDN with existing approaches based on comprehensive simulations and analysis. The results show that our proposed design outperforms previous systems in the service quality and scalability. PACDN has been implemented as an Internet live video streaming service and it was successfully deployed for broadcasting many important live programs in China in 2007. The industrial experiences prove that this design is scalable and reliable. We believe that the wide deployment of PACDN and its further development will soon benefit many more Internet users.