【作者】 冯健；

【导师】 房鼎益；

【作者基本信息】 西北大学 ， 计算机软件与理论， 2008， 博士

【摘要】 音频、视频内容分发代表了Internet中一类重要的应用,但目前还没有满意的解决方案。传统的C/S模式中服务器很容易成为系统瓶颈,IP组播由于难以实现可靠组播和拥塞控制等限制短期内难以在Internet上得到广泛实施,而内容分发网络(content delivery networks,CDN)因费用昂贵而难以推广。研究和应用实践表明,P2P(Peer-to-peer,对等网)流媒体技术能够利用普通节点的资源为其它节点提供服务,在不改变现有网络配置的前提下具有良好的性价比,因而是一种具有广泛应用前景的流媒体分发方法。但由于P2P流媒体系统中Peer节点服务能力的异构性、节点的动态性以及流媒体本身的特殊性,如带宽资源占用高、服务持续时间长、对播放的时限和顺序有严格要求等,使得P2P流媒体分发技术面临诸多挑战。针对大规模流媒体分发服务需求,提供具有高扩展性和QoS(Quality of Service,服务质量)保障的P2P点播流媒体分发服务,已经成为当前流媒体分发技术研究中的一个重要课题。本文围绕上述背景对P2P点播流媒体分发技术进行了深入研究,并提出了有效的解决办法。首先针对QoS敏感的覆盖网络构建问题,提出了一种两阶段节点选择算法TSPS(Two-Stage Peer Selection algorithm)和一个基于多解析环结构的RNLS(Ring-based Network Location Service)节点定位方案。其中TSPS算法在选择提供节点时同时考虑了保证数据可用性和流媒体QoS,即根据播放位置查找候选提供节点后,再根据节点的网络临近性、可用带宽和丢包率等QoS度量参数筛选出合适的提供节点。RNLS节点定位方案采用了半径指数增加的多解析环结构来快速定位节点,有效支持VCR(Video Cassette Recorder)操作。节点间的信息交换采用了基于闲谈的方式。由于节点的服务能力有限,在实际应用中往往需要多个节点才能为单个节点提供流媒体数据服务。设计一个多源流机制的基本挑战是节点的可用带宽未知,而且可能在一个会话期间发生很大的变化。考虑到节点的异构性,本文研究了如何在多个提供节点之间最优地进行带宽分配,其目标是使在播放期限之前收到的数据块的优先级最高。为此分别提出了基于纯拉的调度算法MSS(Multi-SupplierScheduling)和基于推-拉结合的调度算法PPMSS(Push-Pull Multi-Supplier Scheduling)。视频文件通常都很大且对传输有紧迫的QoS要求,很多对等节点可能无法存储整个视频文件以应对网络抖动并为其它节点提供服务。本文形式化了P2P视频点播系统中服务质量敏感的复制问题,其目标是在满足所有节点数据获取最后期限的同时使请求块的成功率最大。为此提出了一个新的缓存机制,包括一个前摄算法和一个缓存替换算法。特别地,定义了数据块的紧迫等级De和贡献度Dc,并将这两个指标用在提出的前摄算法和缓存替换算法中以确定应该请求或替换哪些块。P2P流媒体系统的free riding现象会带来系统性能的严重下降。本文针对P2P流媒体网络中的free riding问题提出了一个基于响应度的激励机制GBIM(Responsivity-Based Incentive Mechanism),其目标是根据节点的贡献提供差别服务,同时最大化系统效用。GBIM使用一个两阶段排序法对请求节点进行排序,并依据顺序优先分配带宽。通过对仿真试验结果的比对分析,表明上述算法和机制性能良好。本文的研究工作为推动P2P点播流媒体的发展与应用提供了良好的理论和实践参考。更多还原

【Abstract】 Audio and video stream distribution is a kind of important application in Internet, there is however no efficient solution. The server often becomes the bottleneck in a traditional client-server streaming system; the deployment of IP multicast is still limited due to several fundamental concerns such as hard to implement reliable multicast and congestion control; and the Internet content providers can not afford the expensive charge of CDN. Because P2P streaming media can provide service to others with one peer’s bandwidth without changing the current Internet deployment; it contains great potential, has wide application foreground. On the other hand, the service capabilities of peers are limited and heterogeneous, and peers can join or leave the system at any time, at the same time, streaming media is a special application with stringent resources requirements for its long last time and playback deadlines, all these make P2P streaming media system face a lot of challenges. Therefore, how to deal with these challenges and guarantee peers’ QoS have been important tasks in P2P on-demand streaming media application. The dissertation studies these problems thoroughly and gives some efficient solutions.According to the problem of QoS-aware overlay construction, a two-stage peer selection (TSPS) algorithm and a peer location scheme based on multi-ring (RNLS) are proposed. TSPS algorithm selects suppling peers according to their cache states and steaming QoS, that is, the first stage of TSPS determines available peers according to peers’ playout position and the second stage determines the most suitable peers according to some QoS metrics such as peers’ bandwidth, delay and loss rate. RNLS is a distributed membership service, in which each peer maintains a data-exchange-ring to explore appropriate data suppliers and several playback-skip-rings with power law radius to assist the quick relocation of VCR operations. Gossip-based method is adopted to exchange message between peers.Because of limited service ability of peers, multiple contents peers are required to be synchronized to send packets to a request peer. The primary challenge in design of a multi-source streaming mechanism is that available bandwidth from each peer is not known a priori, and could significantly change during a session. This dissertation investigates the problem of how to optimally allocate the media streaming among the multiple suppliers, the goal of which is to maximize the priority sum of blocks received ahead of their deadline under heterogeneous bandwidth constraints. Two algorithms are proposed to archive this goal: Multi-Supplier Scheduling (MSS) algorithm based on pull and Push-Pull Multi-Supplier Scheduling (PPMSS) algorithm based on push-pull.Since videos are typically large and have stringent QoS requirements for delivery, many peers may be unable to cache them as a whole to overcome the problem of network jitter and serve others. The QoS-aware replication problem for P2P on-demand systems is formally specified, the goal of which is to satisfy access time deadlines for all nodes while maximize the probability of the successful block requests. A novel caching mechanism is proposed to achieve the goal, which includes a prefetching algorithm and a cache replacement algorithm. In particular, emergency level De and contribution degree Dc of blocks are defined, and used in both algorithms to decide which block should be requested or removed.Prevalent free riding in P2P systems can bring significant degradation in performance. In this paper a P2P VoD incentive mechanism GBIM is proposed, which discourages free riding by letting peers favor uploading to other peers who have proven to be good uploaders according to its responsivity as well as upload bandwidth. Simulation results indicate that GBIM can significantly improve system utility.Through simulation and comparison, performances of both algorithms are illustrated well. Results of our research work will provide a good theoretical and practical reference for P2P on-demand streaming media applications in the future.更多还原