Research On Technologies Of P2P-based Multi-party Video Conferencing

Multi-party video conferencing(MPVC) is the next big challenge for the P2 P streaming technology. For a long time, due to its stringent bandwidth and latency requirements, MPVC has been limited to business market with dedicated equipment and network settings. Commercial MPVC solutions can be classified to two categories, largely server-based and Simulcast.Both of them have performance limitation. Existing academic works study the problem from a theoretical perspective, and are potentially flawed in certain practical aspects. Also, none of them supports any-view mode. We design an overlay network for practical P2 P any-view video conferencing. Through high utilization of network resource, the designed system could optimize user experience, which is important for the development of MPVC.Firstly, this thesis analyzed and described the related technologies of P2 P streaming and multi-party video conferencing. We introduced the characteristics and advantages of P2 P technology, analyzing two main solutions of P2P-based video streaming content distribution. We pointed out the difficulties of how to combine P2 P technology with multi-party video conferencing and analyzed multi-rate support solution which is necessary for dealing with the inherent heterogeneity among streaming users.Secondly, this thesis designed the general architecture and important building blocks for P2P-base multi-party video conferencing. A dualistic topology was designed for the overlay network: a full-mesh for information exchange, and a mesh for data delivery. We use a unified framework to support three multi-rate approaches. To solve the resource competition problem, we assigned different priorities among the sub-layers and versions.Thirdly, the peer design was discussed. We designed multiple key data components and modules for the system of the node. We design a priority based queue scheduling to provide multi-support for users.Finally, the thesis evaluated the different aspects of our system, using a packet-level event-driven simulator written in C++. We demonstrated that our system can deliver close to optimum performance under various stable and dynamic network environments.