A Scalable P2P Video Conferencing System Based on VCStream Model

摘要

Recently, Peer-to-Peer (P2P) based multi-party video conference systems have been widely deployed. However, those solutions suffer from high cost when multiple users simultaneously distribute video streams to receivers. Most existing systems only support up to a dozen of users at the same time. In this paper, we propose a novel approach to construct and manage a heterogeneous P2P network for scalable video conference. First, we design a new hybrid distribution model named VC Stream. It employs Tree-Model (i.e, high performance, real-time, and manageable) and random topology, and hence make the solution high performance, real-time, manageable, scalable and robust. Then we propose a passive algorithm to automatically deal with the nodes' dynamic join and leave in different conditions. After that, according to the four kinds of states of nodes' data buffer, we propose a Push-Pull algorithm to optimize the data transmission. We evaluate our approach in large-scale simulations. The results show that our approach significantly outperforms existing P2P based multi-party video conference solutions with a more scalable, efficient, and consistent performance.