【作者】 郭红方；

【导师】 王宗敏；

【作者基本信息】 郑州大学 ， 计算机软件与理论， 2012， 博士

【摘要】 视频点播VoD (Video-on-Demand)提供的交互式体验改善了人们在传统信息服务的被动模式,以其内容丰富、操作简单、获取便捷等特征得到了飞速发展,己成为当前媒体内容传播的主流方式。然而日益增长的用户规模、海量的媒体数据及高度动态的用户请求,导致现有的内容分发模式已经不能满足用户交互性的需求,即使是当前最流行的P2P分发模式也因用户的高度动态性制约了VoD大规模的应用。在此背景下,研究新的视频点播内容分发技术,实现以低成本向大规模的用户提供稳定而又持续的高质量的媒体服务尤为重要,具有重大的现实意义。针对视频点播系统的数据实时性、节点互助性、资源请求异步性、交互操作等特征,本文对流媒体点播内容分发的关键技术问题进行了深入研究,主要从以下几个方面展开：(1)结合云计算,提出新型的视频点播内容分发服务模式；(2)探讨当用户请求数据片段时,高效快速、准确的服务定位机制；(3)研究利用用户兴趣的局部性(也称空间局部性)提高节点相互共享效率、提升系统服务容量的缓存管理策略；(4)研究数据内容有用性的计算方法,以提高数据缓存的命中率,满足用户流畅跳转的需求；5)建立视频点播应用与云平台交互模型。本文工作主要包括以下几方面：视频点播系统中,用户行为的高度动态性使得支持流媒体快速传输的P2P模式不能满足用户对视频播放的需求。为此,本文提出了结合P2P与云计算的新型视频点播服务模式CAPMedia,从参与节点类型来看,分为P2P端节点、云端节点及源服务器节点。对于终端用户,提出了分布式的双向链表式邻居节点管理策略,不仅有效地改善了频繁更换邻居的不足,也自适应地选择出稳定性较好的节点。用户兴趣相似性与地理位置感知的结点组织策略提高了节点间协助机率,增加了整个系统的服务容量。同时根据P2P点播服务特征,探讨了视频的数据结构模式及性能评价模型。高效的服务定位是视频点播系统设计中一个关键因素,当前的解决方案大多假设用户访问模式服从均匀分布,未能充分考虑用户的自主行为。本文通过形式化对等网络用户搜索模型,分析其影响性能的因素,根据greedy策略得到最优解决方案,并提出一种新的用户行为特征自主发现的自适应服务定位网络DSplay。该结构根据用户的实时访问行为自适应地组成行为相似团体,不需要历史访问记录,就能实时地反应用户的访问模式与数据块的流行度。实验结果进一步验证了DSplay提高了用户服务定位效率,尤其能有效地支持不可预知的非均匀的用户访问请求。预存机制在流媒体点播应用起着关键性作用,随着节点存储能力与网络带宽的不断增加,利用多余的带宽提前获取将来可能需要的视频片段是一个理想的服务模式。本文提出的基于节点离开、邻居协助、流行度感知的预存策略,优化了数据调度策略。与现有预存策略不同,该策略并不预测用户的行为,以一新思路达到降低了跳转时延的目的。通过挖掘搜索结构所学习到的信息,设计了流行度感知的数据预存策略,并通过形式化的分析提出了基于邻居节点离开的优化模型,给出了该问题的精确算法。同时为了使节点间资源共享最大化,进一步提出了节点预存邻居节点中与所请求预存数据片段相关度高的数据片段的策略。在满足用户连续播放需求的前提下,提高了资源共享效率与带宽利用率。视频点播系统中,由于节点访问时间局部性、动态性或其它网络应用,导致用户相互共享已经不能满足其服务容量的需求时,其扩展性受到威胁,然而部署多的服务器导致成本过高。为了提高系统的可扩展性,根据提出的CAPMedia服务模式,探讨了P2P与云计算结合在视频点播应用中的必然性与可行性,对视频点播应用与云平台的交互进行了设计,并提出了考虑节点离开与数据片段流行度的数据迁移策略。对引入云模型服务成本进行了形式化描述,给出一量化的成本估算模型,以指导采取最优的租用策略。最后通过实验验证了所提出策略的有效性,及P2P与Cloud结合的高效性。本论文研究得到国家“863计划”专项课题(2008AA01A315),国家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)及河南省科技创新人才计划项目“P2P VoD关键技术问题研究”(2011HASTIT003)的资助。更多还原

【Abstract】 VoD (Video-on-Demand) supporting interactive operations betters the traditional passive information service, with its characters of rich-content, easy to operate and convenient to fetch. VoD has boomed recently and is becoming the main way of streaming delivery. However, with the growing user scale, huge volume multimedia data and high dynamic of user, the existing content delivery solutions cannot satisfy the interactivity requirement. The large-scale application of VoD remains to be restricted by the dynamic of users with popular P2P technology. Driven by this demand, it is becoming more and more important to improve the quality and efficiency of content delivery and design high quality VoD with low cost to large-scale users.Based on the features of real-time, peer cooperation, asynchronous request, interactivities, the key technologies of Video-on-Demand content distribution are studied intensively from the following several aspects:(1) A novel VoD content distribution service model with collaboration of P2P and Cloud.(2) Effective, quick, accurate, service location mechanism;(3) Cache management scheme which can improve the cooperation of peers and develop the system capacity by analyze user’s interests locality (called spatial locality);(4) A usefulness estimation model for video clips, which improves the cache hit ratio, satisfies user’s requirement of fluent video playback after VCR operations;(5) The interplay model of VoD application and cloud platform. The major contributions are as following.The high dynamics of VoD users restrict the quick streaming delivery with P2P. In chapter Ⅲ, a novel VoD service model (CAPMedia) combining P2P and Cloud is proposed, which includes P2P-assisted peers, cloud-assisted peers, and source servers. In order to address the organization of peers, the neighbor management with Double linked list betters the performance of changeable neighbors; meanwhile, it gives a clue about stable peers’selection policy. The peer management scheme, which is based on the interest similarity and geography-aware features, improves the cooperation of peers and increased system capacity effectively. In addition, the video structure and performance evaluation model are discussed.Efficient service location is a key issue in VoD system, the existing schemes mainly focused on user’s uniform access pattern. In this paper, we formalized the indexing model, carefully analyzed the issues affecting VoD performance, and proposed the optimum solution with greedy scheme. Then a novel adaptive indexing structure is proposed, called DSplay, which is based on user’s arbitrary behaviors, DSplay groups users according to user’s real-time request. It can reflect user’s request pattern and chunk popularity without any history records. The simulation validates the high service location efficiency, especially in supporting unpredictable non-uniform user access pattern.Pre-fetching is a key issue in VoD service, with the growing computer storage and bandwidth capacity, it would be an ideal service model which fetches the video requested in future with idle bandwidth. A novel pre-fetching scheme with peer leaving behavior, neighbor-assisted, popularity-aware, is proposed to futher reduce the latency. It is not nessesary to predict the user behavior which is differs from the existing prefetching schemes. The knowledge extracted from DSplay is used to study a popularity-aware pre-fetching mechanism, then a neighbor assisted optimized model is presented by formal analysis. Moreover, a pre-fetching policy allowing user pre-fetch the video clips which are high related with target clips and are stored by neighbors, is discussed. It improves the sharing efficiency and bandwidth utility with supporting users’ continuous playback efficiently.When users’ mutual sharing cannot support enough capacity for requests due to users’access time locality, dynamic behaviors and other Internet applications, the scalability of system is threatened. It will bring high cost by deploying servers. How to take advantage of reliability, excellent storage and computing ability of cloud with CAPMeida to support VoD application is studied in chapter VI. Based on the analysis of the necessity and feasibility, the interplay between VoD and cloud is designed. Chunk migration policy with peer lifetime and popularity is designed and a cost evaluation model is formalized. Finally, the simulation results show the efficiency of integration of P2P and Cloud.This work is supported by National "863" Project (2008AA01A315),2011 Specialized Research Fund for the Doctoral Program of Higher Education, under grant of "Research of Key Technology of P2P VoD Based on Cloud"(20114101110007) and2011Innovative Talent Project of Department of Henan education, under grant of "Research on key Technology of P2P VoD"(2011HASTIT003).更多还原