【作者】 张茜；

【导师】 王宗敏；

【作者基本信息】 郑州大学 ， 计算机软件与理论， 2014， 博士

【摘要】 视频点播VoD（Video-on-Demand）实现了用户与视频资源高度交互的访问模式，使其得到越来越多用户的青睐，成为目前互联网上需求量最大的服务之一。点播服务中用户交互式操作、用户实时性观看需求、服务规模爆炸性增长及用户视频数据请求量不稳定性等特点给实现一种高用户满意度且经济有效的点播服务带来了一定的问题。考虑到点播系统中视频片段及视频间存在的关联性特征，本文基于关联性特征对如何实现经济有效且高质量的点播服务进行了优化研究，主要从以下几个问题进行展开：1）快速资源定位机制的实现。针对该问题，提出一种基于视频片段关联性的点播服务模型。所提模型利用视频点播中视频各个片段之间关联性强度不同这一特点，使具有相似兴趣点的节点能够自主地组织在一起，从而有效地提高了节点在进行转跳操作时查找目的片段的效率，提升了用户观看满意度。对由该策略引出的节点邻居优化问题进行了分析，将其形式化描述为集合覆盖问题，并提出一种基于贪心算法的邻居近似优化策略来解决该问题。仿真对比实验验证了所提策略在降低查找时延方面的有效性，并对所提邻居优化算法的有效性进行了验证。2）有关联性视频的点播服务在云平台上的部署。针对该问题，本文通过对YouTube上有关联性关系视频的采集，得出用户以近似90%的概率会选择当前视频三跳关联以内的视频进行观看这一结论，基于此提出一种基于簇的P2P云辅助点播服务模型。在所提模型中，由于云服务器存在一定的启动延迟，因此对簇的带宽迁移量进行了预测；针对云服务器有着不同的类型、租用代价以及服务能力的特点，提出一种满足用户带宽需求前提下降低云服务器租用代价的云服务器租用策略，仿真对比实验验证了所提策略在降低服务资源开销以及提高用户满意度上的有效性。3）关联性视频内容在分布式云服务器上的内容放置问题。针对该问题，考虑到关联性视频的区域流行度差异，提出一种地域感知的关联性视频内容在分布式云服务器上的内容放置策略。该策略目的是在尽量维持关联性关系及不同云服务器上负载均衡的同时降低用户跨地域请求率。将问题形式化描述为一个有限制的K中心点划分问题，提出一种地域感知的PAM算法解决该问题。仿真对比实验表明了该算法在提高用户本地请求率和降低用户跨区域请求方面的有效性，同时也能较好地维持视频之间的关联关系。4）提高节点缓存空间利用率的关联性视频内容缓存替换问题。针对此问题，提出一种基于视频相似的缓存替换策略。所提策略优先考虑替换同已替换视频集合语义相似度较大即可能为不受欢迎的视频，并使替换视频整体流行度尽可能地小且副本数尽可能地大。该问题可描述为一个多目标优化问题，将其转换为单目标优化问题并进一步给出缓存替换策略。仿真实验分析了策略中参数设置对缓存命中率的影响，并证实了所提策略在提高缓存内容命中率上的有效性。本论文研究得到国家“863计划”专项课题（2008AA01A315），教育部高等学校博士学科点专项科研基金课题“基于云服务的视频点播关键技术研究”(20114101110007)，河南省科技创新人才计划项目“P2P VoD关键技术问题研究”（2011HASTIT003)以及河南省教育厅重点项目“基于云平台的P2P VoD关键技术研究”(13A520562)的资助。更多还原

【Abstract】 VoD (Video-on-Demand) supports interactive operations, is better than thetraditional passive video service and is becoming one of the most popular services inthe internet. The characteristics of VoD, such as the interactive operations, user’sdemand for the real-time view, the explosive increase of service scale, as well as theinstability of users’ demand, bring some challenges to implement a high-quality andcost-effective VoD system. This paper, basing on the relevance characteristic ofdifferent video segments and videos, implements an optimizing research on the keyproblems of VoD service which are as following:1) Fast locate the destination segment. To solve this problem, based on thecharacter that the segments of on-demand streaming have some relevance with eachother, a new VoD model is proposed. The proposed model makes peers with similarinterests organized together to realize most users’ requests can be satisfied by theirneighbors. And then it improves the searching efficiency of locating the destinationsegment. The optimization problem about how to optimize peer’s neighborlist isinduced by the proposed strategy, and the problem can be formulated as a set coverproblem. An approximate optimization algorithm is presented to solve the problem.The simulation results show that the proposed strategy can effectively reduce theseeking delay and enhance the scalability of the system.2) The deployment of VoD service on cloud platform. Through the crawled dataof videos in YouTube, we find that with larger than90%probability, the YouTubeuser’s all requested videos are within three hops of related videos. Base on thisconclusion, a cluster-based P2P VoD model with cloud assistance is proposed. Therequested bandwidth prediction for a cluster is needed for the start delay of cloudservers. Given the diverse capacities, cost, limited lease size of cloud servers, weformulate an optimization problem about how to lease cloud servers to minimize theleasing cost. And then a heuristic solution is presented. The evaluation shows theefficiency of the proposed schemes. 3) The research on the optimal content placement scheme on distributed cloudservers. Considering the regional differences in the videos’ popularity, ageographic-aware content placement scheme is proposed. The proposed scheme aimsat reducing the cross-boundary traffic and realizing the load-balance on cloud serversas well as preserving the social relationship. The problem can be formulated as aconstrained k-medoids clustering problem which under the constraint of minimizingthe cross-boundary traffic and the imbalanced weight on cloud servers. Comparedwith the previous scheme which overlooked the geographic popularity of interest, theproposed one effectively reduce the cross-boundary traffic and realize theload-balance on cloud servers as well as preserve social relationship.4) The research on an efficient cache replacement strategy to promote thecontent utilization and reduce the delay of content request with the limited cache size.This paper proposes a cache replacement scheme based on video semantic similarity.When a peer’s cache space is full, it will replace the videos which have a lagersemantic similarity with the already replaced videos and the replaced content shouldhave the smaller popularity and larger replications. The replacement problem can bedescribed as a multi-goal optimizing problem, we transform it into a single-goaloptimizing problem and then describe it as a knapsack problem. A heuristic algorithmis proposed to solve it. The simulation analyses the impact of different parametersettings on the performance of the proposed scheme and verifies the effectiveness ofthe scheme in promoting the hit ratio.This paper is funded by National “863” Project (2008AA01A315),2011Specialized Research Fund for the Doctoral Program of Higher Education, undergrant of “Research of Key Technology of P2P VoD Based on Cloud”（20114101110007） and2011Innovative Talent Project of Department of Henaneducation, under grant of “Research on key Technology of P2P VoD”(2011HASTIT003), Key Projects in Henan Province Department of Education, undergrant of “Research on Key Problems of Cloud-assisted P2P VoD”(13A520562).更多还原