Peer Streaming Media On Demand System Scheduling Mechanism And Caching Strategies

The peer-to-peer video-on-demand streaming system, because of its good interactivity, excellent system scalability and low deployment cost, has become the major direction of Internet video application. At present, designing and deploying a peer-to-peer video-on-demand streaming system with better user experience, guaranteed quality of service and less server stress still face some open problems which are requiring further research. In these open questions, the scheduling mechanism and caching strategy are two major aspects which have a significant impact on the system performance. Although at present there is much research work focusing on scheduling mechanism and caching strategy and some practical system successfully deployed in the Internet, these existing schemes still have some limitations respectively and enjoy large space for optimization.After deeply researching of the data scheduling mechanism and caching strategy of peer-to-peer video-on-demand system, this thesis proposes a hybrid push-pull scheduling mechanism with random linear network coding in Galois field 2 and a weight function based caching strategy. What's more, a prototype system according to these schemes is designed. The major contributions of this thesis can be concluded as follows:1. A hybrid push-pull scheduling mechanism with random linear network coding in Galois field 2 is proposed. After careful and suitable segment and block division on video, random linear network coding in Galois field 2 is carried out on the blocks within a segment, and then the push-pull hybrid solution is used to schedule the blocks and segments. This solution not only remarkably reduces the complexity of the coding, greatly improves the collaboration between peers, but also reduces the redundant data as well as the system latency.2. A weight function based caching mechanism is constructed which considers not only the emergency degree, rare level but also the association rules mining results of user jump behavior. The association rules mining of user jump behavior is introduced to reveal the characteristics of the video itself to some extent. And a prefetching algorithm and caching algorithm are proposed according to the weight function. In this strategy, both the individual requirements of single peer and the impact on the overall performance of the system are considered. So the limited buffer space can be better used and the system performance can be improved.3. A prototype system design based on our schemes is given out including a system topology design, system architecture design and peer's module structure design. At the same time, a simulation tool is carefully implemented to evaluate the performance of the schemes proposed in this thesis, using the existing real measurement results from Internet such as end-to-end delay matrix, heterogeneous bandwidth capacities of peers and user behavior modes. The performance indexes to be investigated include playback continuity, startup latency, jump latency, server stress, dependence of network coding and system control overhead and so on. The simulation experiment results show that, in our schemes, the system can achieve better scalability and playback continuity, lower system latency especially jump latency. Thus, the VCR operation can be well supported and the system performance can be optimized.