| Multimedia streaming over the Internet has had large success and popularity mainlydue to the booming applications, e.g., Sopcast, PPlive and Youku. These applications aremostly based on Client/Server (C/S) mode and consume a huge amount of resources interms of server bandwidth and storage. To overcome these limitations, many Peer-to-Peer(P2P) streaming systems (e.g., PPLive, PPS) have been developed to realize scalable andcost-effective content distribution. An important issue in P2P streaming system design ishow to assign limited available resource for a large number of peers to ensure dataavailability and further to improve peers’ startup delay and playback continuity.In this thesis, a fluid-flow content distribution model is presented to describe indetail chunk distribution in P2P streaming system. The proposed model regards the chunkdelivery process in a peer-assisted setting. Chunks are ordered independently and raise therivalry among peers due to the race of getting chunks as early as possible. We proposednew leechers join the system if there is enough bandwidth available for them: nodes inupload state can share all chunks in their memory, with new leechers. To manage this oldpeer closed to the server can fetch the chunks from the server given that the competitionfor the server is high and thereafter they will assist the server like sharers by contributingtheir resources to other peers; the system will maximize the number of sharers that keepsmooth playback continuity.This strategy promises to reduce unwise bandwidth allocation to admit new leechers,when existing peers help to achieve excellent VoD performance. And content distributionmodel we propose here prove that the random exponential backoff scales well with thetotal number of peers, and can obtain near-optimal distribution of chunks. Our model andparameter estimation are meaningful for designers and operators to improveQuality-of-Experience metrics of P2P streaming systems. |
PDF Full Text Request