Research On The Key Technologies Of Resource Allocation For P2P VoD

With the growing popularity of broadband Internet and the rapid development ofmultimedia technology, the network applications have gainedrapid growth andenrichment in recent years.Among them, the Video-on-Demand (VoD) has attracted agreat number of users and become a killer application over the Internet due to itsrich-content and convenience to fetch. With the tremendous volume of multimedia dataand growing number of users, the construction of a highly efficient distribution systemwith low deployment cost has become one of the most urgent needs for VoD service.Peer-to-Peer (P2P) technology brings a new opportunity for the development of VoDservice. Studying how to build a high-performanceVoD system by utilizing thebandwidth resource of peers in the P2P network has become a hotspot of currentacademic research. However, the lack of synchronization among users, the requirementfor playing video smoothly and the peer dynamics bring a great challenge for P2P VoD.In view of the above, this thesis focuses on the study of resource allocationmechanisms for P2P VoD systems. Several reasonable analytical models and efficientoptimization technologies of resource allocation for P2P VoD are proposed, aiming atimproving system performance and reducing inter-ISP traffic of the network byeffectively allocating bandwidth resource of peers. Main contents of this thesis are listedas follows:1) Proposing a Load Model based on the server load to evaluate the intra-channeland inter-channel bandwidth imbalances in the multi-channel P2P VoD systems. Byintroducing the queuing network theory, this model can capture several aspects of peerbehavior, such as participating in the system, sojourning in a channel, wandering aroundchannels and leaving the system. The average load index is defined as a performancemetric to evaluate the performance of three potential bandwidth allocation designs forP2P VoD: Naive allocation (NAA) design, Independent-Channel Allocation (ICA)design and Cross-Channel Allocation (CCA) design. By developing an asymptotictheory to provide theoretical results, this thesis analytically shows that NAA incurs boththe intra-channel and inter-channel bandwidth imbalances seriously; ICA can reduce theintra-channel bandwidth imbalance but still incurs the inter-channel bandwidthimbalance; and CCA can reduce both of the two bandwidth imbalances. Then, accordingto the theoretical analysis, this thesis proposes CCA(Z), a optimization strategy based on the CCA design. The idea of this strategy is that each leecher (i.e., downloading peer)selectsits neighbors sequentially in the currently viewing channel to reduce theintra-channel bandwidth imbalance and each seed (i.e., seeding peer) serves consecutiveZvideo chunksfor a selected channel through cross-channel collaboration, aiming atreducing the server load. The simulation results not only validate the theoretical analysis,but also show that the CCA(Z) strategy outperforms the state-of-the-art approaches interms of server load.2) In order to alleviate server load, reduce inter-ISP traffic and increase peerprefetching, this thesis proposes two ISP-aware centralized resource allocationalgorithms: Server-First Optimal (SFO) algorithm and Locality-First Optimal (LFO)algorithm. SFO algorithm takes minimum server load as the first order objective, andthen minimizes inter-ISP traffic and maximizes peer prefetching sequentially. LFOalgorithm takes minimum inter-ISP traffic as the first order objective, and thenminimizes server load and maximizes peer prefetching sequentially. By introducingnetwork flow theory, SFO and LFO can be solved as a series of maximum or min-costflow problems. Simulation results show that the two proposed algorithms substantiallyoutperform the traditional and state-of-the-art approaches.3) Given that the SFO and LFO algorithms are not suitable for dynamicenvironment very well, this thesis proposes Linear Weighted Sum with HybridNeighbor (LWS-HN)mechanism, an ISP-aware distributed resource allocationmechanism for P2P VoD. LWS-HN utilizes a novel hybrid neighbor selection approachwhich considers both peer load balance and network locality, and models the tradeoffsamong server load, inter-ISP traffic and peer prefetching as a multi-objectiveoptimization problem whose objective function is the linear weighted sum of the threeobjectives. A distributed method of solving the optimization problem is proposed,whichcan be converted into a distributed rate allocation strategy for P2P VoD. LWS-HNmechanism can alleviate server load, reduce inter-ISP traffic and increase peerprefetching simultaneously or achieve the tradeoffs among them if there is any conflict.Simulation results show that LWS-HN mechanism substantially outperformsthetraditional and state-of-the-art approaches.It is of higher practical significance and theoretical value for the research of thisthesis. This study could not only provide fundamental insights into resource allocationfor P2P VoD, but also give helpful reference for improving performance and reducingdeployment cost.