P2P点播资源分配关键技术研究

发布时间：2018-01-22 03:13

本文关键词： 对等网络视频点播流媒体理论建模资源分配　出处：《重庆大学》2014年博士论文　论文类型：学位论文

【摘要】：近年来，宽带互联网的普及和多媒体技术的发展带来了网络应用的不断增长和丰富。其中，视频点播(Video-on-Demand，VoD)业务以获取便捷、内容丰富等特征吸引了广大的用户，己成为当今互联网上的杀手级应用。海量的媒体数据和日益增长的用户数量使得以低成本构建高质量的分发体系成为视频点播业务中最迫切的需求之一。P2P技术为视频点播业务的进一步发展提供了新的契机。研究如何利用P2P网络中的节点带宽资源来构建高性能的视频点播系统是当前学术界的热点。然而，点播用户行为的不同步、对视频流畅性的需求以及用户动态性等因素给P2P点播带来很大的挑战。本文针对P2P点播系统中的资源分配技术展开深入研究，针对现有技术的不足，提出几种合理的资源分配模型和高效的资源优化技术，旨在通过合理有效地分配节点的带宽资源，达到提高系统性能、减少跨ISP流量的目的。本文的研究工作主要包括以下三个方面： 1)提出一个基于服务器负载的模型来研究多频道P2P点播系统中的频道内带宽失衡现象和跨频道带宽失衡现象。该模型引入排队网络理论，能够较好地描述节点的各种行为特征，比如加入系统、在频道中逗留、切换频道和离开系统等。该模型定义平均负载指数作为性能评价指标，进而分析和比较了三类可用于多频道P2P点播系统的带宽分配方法，即朴素分配(NaiveAllocation，NAA)方法，，独立频道分配(Independent-Channel Allocation，ICA)方法和跨频道分配(Cross-ChannelAllocation，CCA)方法。通过渐进分析的数学方法，在理论上证明了NAA方法将同时导致频道内带宽失衡和跨频道带宽失衡；ICA方法能够有效减轻频道内带宽失衡但依然会导致跨频道带宽失衡；而CCA方法能够同时减轻频道内带宽失衡和跨频道带宽失衡。在理论分析的基础上，提出一种基于CCA方法的优化策略CCA(Z)。该策略的核心在于正在下载的节点在频道内按顺序选择邻居以缓解频道内带宽失衡，种子节点以跨频道资源共享的方式来服务选中频道的连续Z个媒体块；这样做能够达到降低服务器负载的目的。通过仿真实验，不仅证明了理论分析的正确性，也证实了CCA(Z)在服务器负载方面具有比现有技术更好的性能。 2)为达到降低服务器负载、减少跨ISP流量和提高节点预读取总量这三个目标，本文提出两种感知ISP的集中式资源分配算法。一是以最小化服务器负载为首要目标、最小化跨ISP流量为第二目标、最大化节点预读取总量为第三目标的服务器优先优化(Server-First Optimal，SFO)算法；二是以最小化跨ISP流量为首要目标、最小化服务器负载为第二目标、最大化节点预读取总量为第三目标的位置优先优化(Locality-First Optimal，LFO)算法。通过引入网络流理论，这两种算法可以转化为一系列的最大流问题或最小费用流问题进行求解。仿真实验证明，SFO和LFO算法拥有比现有技术更好的性能。 3)针对SFO和LFO算法不能很好的应用于动态环境的问题，本文提出基于混合邻居的线性加权和(Linear Weighted Sum with Hybrid Neighbor，LWS-HN)机制。这是一种感知ISP的分布式资源分配机制，采用一种兼顾节点负载均衡和流量本地性的混合邻居选择策略，并通过构造一个目标函数是服务器负载、跨ISP流量和节点预读取总量三者的线性加权和的优化问题来进行节点的速率分配。针对该优化问题提出一种分布式的求解方法，并将其转化成一种能够用于P2P点播系统的动态速率分配策略，从而实现减轻服务器负载、减少跨ISP流量和提高节点预读取总量这三个目标或在这三个目标出现冲突时达成某种平衡。仿真实验证明，LWS-HN机制的性能表现优于现有技术。本文的研究具有较高的现实意义和理论价值，一方面可以对P2P点播中的资源分配技术提供深入的认识，另一方面可以对提高系统性能、降低部署成本提供有价值的参考。
[Abstract]:In recent years, the development of the popularity of broadband Internet and multimedia technology brings the rapid growth of network applications and rich. Among them, the video on demand (Video-on-Demand, VoD) business to get convenient, rich content and other characteristics to attract the majority of users, has become the killer application on the Internet. The number of media data and growing the user makes the distribution system of high quality with low cost has provided a new opportunity for further development of.P2P technology is one of the most urgent needs of VOD service for the VOD service. VOD system on how to use the bandwidth resource in P2P network to build a high performance is a hot topic in academic circles. However, not synchronized user behavior, video on demand and user flow of dynamic factors to P2P demand brought a great challenge.
This paper conducts in-depth research on resource allocation for P2P VOD system, aiming at the defects of the existing technology, the technology of resource optimization resource allocation model of several reasonable and efficient, through the rational and efficient allocation of node bandwidth resources, improve system performance, reduce the cross ISP traffic. The main research work of this paper includes the following three aspects:
1) proposed a bandwidth imbalance load model based on multi channel P2P VOD system to study the channel and cross channel bandwidth imbalance. This model introduces queuing network theory, which can describe the behavior of nodes, such as adding system, stay in the channel, the channel switching and leave the system. This model defines the load average index as the performance index, and then analyze and compare the three classes can be used for bandwidth allocation method for multi channel P2P VOD system, namely simple distribution (NaiveAllocation, NAA) method, independent channel assignment (Independent-Channel Allocation ICA) method and cross channel allocation (Cross-ChannelAllocation, CCA) method. Through mathematical method asymptotic analysis, we prove that the NAA method will also lead to imbalance of bandwidth and cross channel bandwidth imbalance channel method can effectively reduce the ICA; In the channel bandwidth imbalance but still can cause the cross channel bandwidth imbalance; while the CCA method can also reduce the channel bandwidth and cross channel bandwidth imbalance imbalance. On the basis of theoretical analysis, put forward an optimization strategy based on CCA CCA method (Z). The core of this method is under the load of the nodes in the channel at in order to alleviate the imbalance of channel bandwidth choice of neighbors within the seed node to cross channel resource sharing services to selected channel Z continuous media blocks; it can reduce server load. Through simulation experiments, not only proved the correctness of theoretical analysis is confirmed by CCA (Z) in the server load has better performance than existing techniques.
2) in order to reduce the server load, reduce the cross ISP traffic and improve the total node pre read the three objectives, proposed centralized resource allocation algorithm for two kinds of perceived ISP. One is to minimize the server load as the primary goal, to minimize the cross flow rate of ISP is second, the maximum node pre read a total of third goals server priority optimization (Server-First Optimal SFO) algorithm; two is to minimize the cross ISP traffic is the primary goal, to minimize the server load is second, and the maximum node position of pre read total third target first (Locality-First Optimal LFO) optimization algorithm. By introducing the network flow theory, the two algorithms can be transformed into the maximum flow problem of a series or the minimum cost flow problem. Simulation results show that the SFO and LFO algorithm has better performance than existing techniques.
3) for the SFO and LFO algorithms are not well applied in the dynamic environment, this paper puts forward the linear and mixed weighted neighbor based (Linear Weighted Sum with Hybrid Neighbor, LWS-HN). This is the mechanism of distributed resource allocation mechanism of a perception of ISP, using a mixture of both the neighbor node local load balancing and flow the choice of strategy, and by constructing an objective function is the server load, optimization problems and linear weighted cross ISP traffic and three of the total amount of prefetch nodes to rate allocation nodes. For the optimization problem is proposed for a distributed method, and converted it into a can be used for dynamic rate allocation strategy P2P VOD system, thereby reduce server load, reduce the cross ISP traffic and improve the total amount of the three node prefetch target or reach some kind of conflict in the three goals The simulation experiment shows that the performance of the LWS-HN mechanism is superior to the existing technology.
The research in this paper has high practical significance and theoretical value. On the one hand, it can provide an in-depth understanding of resource allocation technology in P2P on demand, and on the other hand, it can provide valuable reference for improving system performance and reducing deployment cost.

【学位授予单位】：重庆大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN948.64

【参考文献】