向大规模用户以可扩展的方式提供多媒体点播服务是一个有挑战性的课题。对等技术通过聚积系统节点的存储带宽资源，为大规模的点播系统设计提供了一个有吸引力的解决方案。基于对等技术的流媒体直播服务在过去几年中成为学术界的研究热点。对等直播服务在实际被成功部署、其关键技术被深入研究。点播服务具有高带宽需求、非同步用户请求、用户互动操作以及内容实时传送等方面的特点，与直播服务相比更复杂。对等网络点播协议还没有被很好的研究，这也是本文研究工作的主要动机。其主要研究贡献如下： I 本文分析了对等网络点播服务的特点，提出了通过播放偏移点来分布式的组织节点的方法。通过构造双尺度列表成员，节点从两个尺度来维护系统视图。在细粒度下，节点与具有相似播放偏移点的S节点构成协作组。在粗粒度下，节点通过维护R节点来支持查找操作，从而在存在互动操作时也能提供点播服务。节点间的协作方式通过蜂群协议来实现，本文考虑到节点间缓存的异步性、数据紧迫性和数据稀有性，提出了需求驱动的蜂群算法，实验结果显示该方法在突发场景下和泊松到达场景下都能大大降低服务负载、保持播放的连续性。 II 本文考虑到对等点播系统中节点异构属性的问题，研究了如何在线检测超级节点。通过将结构化组织超级节点，为后续的点播请求提供服务云。服务云上的内容采用被动管理的方式，该方法在动态的对等网络中能够避免带宽的浪费。为了达到高效更新和定位数据的能力，本文分析了数据的内部关联，在结构网络之上加入逻辑链状结构来管理顺序依赖的数据块集合。由于后台服务器能够快速的定位到足够多的数据块提供者，本文提出了两个负载均衡的数据调度算法。仿真结果验证了该方法能够高效的提供点播服务，具有很好的扩展性。 III 分析了ISP与对等网络应用之间的矛盾，提出了基于动态标记点的位置感知覆盖构造方法。动态标记点方法不需要ISP参与就能够提供节点间的相互近邻关系。通过分裂合并的方式，节点在形成协作组的过程中考虑基于近邻的聚类效应。该方法同样也适用于流媒体直播服务。对具有相同下载目的的节点，给出了一种混合树网结构的组织方式。仿真结果显示该方法不仅达到降低跨ISP流量的目的，同时节点传输效率也得到提升。
英文摘要
Providing video-on-demand(VoD) service to a large number of users in a scalable manner is a challenging problem. Peer-to-Peer(P2P) technology is an attractive solution for scalable VoD application because it can aggregate storage and bandwidth capacity of peers and alleviate workload on the source server. P2P live streaming technology has become hot topic for both academic researchers and industrial engineers; and it has been well studied and successfully deployed recently. Compared with P2P live streaming, it is much more challenging for P2P-VoD due to peers' asynchronous request , different content focus, user interactions and streaming requirement of video contents. P2P-VoD is not well studied until now, this status motivate the work of this thesis. The major contributions of this thesis are listed as follows: I We address the asynchronous problem in VoD system with playback offset model. Peers who join the system randomly are ranked by their playback offset. Therefore, peers with similar offset share overlap content focus, and data blocks can be exchanged among them. In order to locate random peers without centralize server, a distributed two scale list which consists of S-peers and R-peers is proposed. With the help of R-peers, Peers can locate the desired peers when they perform VCR-like operations such as random seek. Considering urgency, rarity of chunks and asynchronous buffer, we propose a sophisticated demand-driven chunk swarming protocol to collaborate with ragged partners(S-peers). This protocol alleviates the workload on source server effectively, and guarantees playback continuity. II We propose a novel P2P-VoD scheme called PCMS. We firstly explore the existing super peers in a VoD session, which are the stable peers with more capacity than normal peers. Contents cached by super peers are organized by a structure network with put and get functions. The content network, which is composed of source server and super peers, acts as the service cloud of the video session. Later peers can retrieve chunks from the cloud. For the sake of utilizing peers' bandwidth efficiently, load-balancing schedule protocol is used among super peers and normal peers. With trace data from real P2P-VoD systems, extensive simulation experiments show the efficiency and scalability of PCMS. Furthermore, PCMS can provide true VoD service with VCR-like interactions. III The conflict between ISP and P2P application is well studied. In order to design an ISP-frien...
修改评论