Research On Key Problems Of Controllable P2P Media Streaming Distribution

Owing to the advantages of good scalability, low cost and easy-to-deploy, the technology of P2P has become a dominant method for media streaming applications in recent years. This dissertation chose media streaming distribution under controllable P2P environments as a starting point, and conducted a sophisticated research on its key technologies after fully analysis and conclusion upon current studies of media streaming. New algorithms and mechanisms are presented and related performances are verified by simulation and some good results are achieved in the dissertation. Main innovations and contributions of the dissertation include the following:1) Researching the evolutionary routines of media streaming distribution and proposing the concept of Controllable P2P. Firstly the evolution routines of media streaming distribution technologies are revealed and key techniques and state-of-arts of different media streaming distribution architectures are briefed. And then the concept of Controllable P2P is proposed by summarizing the newly-emerging media streaming systems. Controllable P2P refers to the Peer-to-Peer distribution technology for controllable networks, in which network operators- controlled equipments, such as set-top boxes, home gateways, take part in the P2P computing. Essentially, Controllable P2P extends the application multicast ability to the edge devices of network infrastructure from previous user-side machines, which can help to overcome the defect of churns in normal P2P technology and imports many outstanding characteristics. It would be a valuable research topic to improve performance of media streaming distribution by Controllable P2P.2) Proposing the helper-assisted altruistic interacting scheme for P2P media streaming in Controllable P2P framework. Problems and algorithms of peer selection, neighbor discovery and data scheduling etc. are fully studied. Compared to existing solutions, this one needs no additional encoding on media data and is totally compatible to the prevailing data-driven mode, which results in an easy fusion into current distribution frameworks. The experimental results show that this altruistic interacting mechanism can effectively relieve server load and make playback more smoothly, and has the series of advantages of scalability, reliability, transparency and so on. 3) Proposing caching replacement model for media streaming which based on global and local supply-demand patterns. Concepts of global supply-demand pattern and local supply- demand pattern are distinguished and distributed consensus algorithm is introduced as a supply-demand solving method. When evaluating cache supplying capacity, it takes both bandwidth heterogeneous and service competition factors into consideration, by which actual supplying ability can be reflected more accurately. In this model local supply-demand pattern is used for cache replacement within area network, while on cache servers both local and global supply-demand patterns are utilized for cache placement and replacement. The following experiment shows that this model can improve data sharing efficiency within area networks, reduce inter-area traffics, as well as decrease server loads in large-scale system.4) Researching internal popularity model of streaming media files. A double-exponential model is proposed for streaming media internal popularity based on analysis of real system log files. The fitting results in experiments shows that this model predicts the internal popularity more accurately than models in other literatures like exponential model, Zipf model and stretched exponential model etc.5) Proposing a transparent proxying service model for tree topology in controllable P2P networks and further researching replica placement problems in this model. By case analysis, it is found that prefix placement can hardly achieve the near-optimal placement performance in tree topology. Supported by double-exponential internal popularity model and using data segment as its replica granularity, a non-fixed length and non-continuous replica placement model is proposed which considers both proxy storage capacity and serving throughput. Two algorithms, HGPA and GGPA are given in this model based on greedy strategy. Simulation results demonstrate that these proposed algorithms can improve replica hit ratio and decrease accessing cost effectively. It can be concluded that non-continuous placement is superior to prefix placement in tree networks.