【作者】 邓光青；

【导师】 陈常嘉；

【作者基本信息】 北京交通大学 ， 通信与信息系统， 2012， 博士

【摘要】 伴随着用户接入宽带的快速增长、个人计算机性能的不断提升以及网民规模的日益扩大,P2P流媒体系统以其巨大的系统容量、稳定的服务质量、良好的可扩展性以及低廉的部署成本,迅速发展成最热门的互联网应用之一,能支持数以百万的用户同时观看高清视频,极大地推动了资讯在互联网上的快速传播。在获得迅猛发展的同时,P2P流媒体系统还存在着诸多等待解决的问题,如服务质量的特性未揭示、带宽资源的分配不合理以及信令开销太大等。针对上述问题,本文从理论建模、网络测量及协议设计这三个方面开展工作,取得的主要研究成果如下：(1)针对P2P点播系统中服务质量与关键参数的数学关系未披露,演播中断的特性未揭示的问题,建立了针对演播中断的服务质量评估模型。该模型首先利用带漂移的布朗运动对节点的下载过程和视频的演播过程分别进行建模,然后建立了演播流畅度与视频码率、节点下载速度以及预取数据量之间的数学关系,最终推导了中断概率、首次中断时间以及中断次数的精确数学表达式。由上述模型可得,下载速度与演播码率的动态性是降低演播质量的主要原因；增大预取数据量或下载速度都可提高演播流畅度,但前者可使中断推迟发生。在视频的演播过程中,起始阶段发生中断的概率很高；在一次演播过程中,发生多次中断的现象是存在的,但其概率很小。为验证播放中断的性质,对PPLive点播系统的数百万个用户展开了为期5天的网络测量且实测结果与上述结论是一致的。上述模型不仅揭示了P2P点播系统的特性,更为新带宽分配算法的设计奠定了理论基础。(2)针对P2P点播系统中由于各节点分得的带宽资源与节点实际需求之间的匹配度低、联动性差(节点分得的带宽资源不能随节点需求量的变化而实时改变)而导致节点演播质量严重分化的问题,建立了一个节点带宽需求模型,给出了节点期望带宽与预取数据量、播放码率以及演播流畅度之间的数学表达式。基于该模型,提出了一种分布式上载带宽分配算法,该算法可以实现系统带宽资源在异构节点之间的实时动态分配,使预取数据量多或所观看视频码率低的节点释放多余的带宽,并将释放的带宽分配给预取数据量少或所观看视频码率高的节点,最终使所有节点都能获得较高的演播流畅度。仿真分析表明,与传统的贪心策略相比,该算法既能缩短节点的启动延时又能降低服务器的带宽消耗。(3)针对P2P点播系统中由于节点无预取限制而导致节点演播质量两极分化的问题,提出了一种基于视频分块的适度预取策略。在该策略中,预取数据量大的节点会主动停止下载并将腾出的带宽分配给预取数据量小的节点,从而使得低下载能力节点也能获得与高下载能力节点同样的演播质量。针对该策略建立了一个性能评估模型,由该模型可得,该策略可使节点获得适度的预取数据量以及较高的演播流畅度。最后,将该策略部署到PPLive点播系统中,实测结果表明,PPLive点播系统可获得高达95%的服务器带宽节省率,而未采用该策略的P2P点播系统UUSee的服务器带宽节省率只有70%。且PPLive点播系统的平均中断次数仅为0.8次,即视频播放非常流畅。总之,基于视频分块的适度预取策略既能降低服务器带宽又能提高节点演播质量。更多还原

【Abstract】 With the increase of user access bandwidth and the number of Internet users, as well as the improvement of PC capacity, P2P streaming media system becomes one of the most popular Internet applications, due to its huge system capacity, stable quality of service (QoS), outstanding scalability and low deployment cost. P2P streaming media system can support millions of users watching high definition video at the same time and then greatly speed up the transmission of information over Internet. Though gaining rapid development, P2P streaming media system still has some pending issues, for instance, the QoS characteristic has not been revealed, the assignment of bandwidth resource between peers is not reasonable and the signal overhead is too high. Towards these issues, we study P2P streaming media system in terms of theoretical modeling, network measurement and protocol design. The main achievement is below:(1) Towards issues in P2P VoD (Video-on-Demand) system that the relationship between QoS and important parameters is not revealed and the playback interruption characteristic is not disclosured, we establish a QoS evaluation model towards playback interruption. In this model, we first use the Brownian motion with drift to model the peer downloading process and video playback process, then establish the mathematical relationship among playback continuity, video playback rate, peer download rate and prefetched data amount (PDA) and finally get the mathematical expression of the playback interruption probability, the first interruption time and the number of interruptions during a viewing process. From above models, it is observed that the dynamics of peer download speed and video playback rate is the main cause of QoS decline; increasing PDA or download speed can improve playback continuity but the former delays the playback interruption. During a viewing process, the interruption occurs at the beginning with great probability; many interruptions may happen with small probability. To verify the playback interruption characteristic, we do a network measurement towards PPLive VoD system with millions of users for continuous five days and the measurement results are consistent with the conclusion derived from the model, which not only reveals the unique characteristic of P2P VoD system, but also lays the foundation for new bandwidth assignment algorithms.(2) Towards the issue that different users in P2P VoD system have different watching experience because the bandwidth obtained by one peer is not consistent with the bandwidth required by that peer (namely the bandwidth obtained by one peer can not dynamically and in real time change with the bandwidth requirement of that peer), we establish a peer bandwidth demand model, which shows the mathematical relationship among expected bandwidth, PDA, playback rate and playback continuity. Based on this model, a distributed upload bandwidth assigning algorithm is proposed. This algorithm achieves real-time and dynamic bandwidth assignment among heterogeneous peers; specifically, this algorithm makes peers with too much prefetched data or lower playback rate release some bandwidth, which then is assigned to those peers with too little prefetched data or higher playback rate. In this way, all peers in P2P VoD system can obtain the same playback continuity. Simulation results show that compared with traditional greedy strategy this algorithm gains shorter startup delay and less server bandwidth.(3) Towards the polarization of peer QoS due to no prefetching limitation in P2P VoD system, we propose a moderate prefetching strategy based on video slicing mechanism. In this strategy, peers with more prefetched data will stop downloading content and the released bandwidth is assigned to those peers with less prefetched data. In this way, low download capacity peers can gain the same playback continuity as the high download capacity peers. Towards this strategy, we propose a performance evaluation model from which it is observed that under this strategy peers can obtain moderate prefetched content and high playback continuity. Finally, this strategy is applied to PPLive VoD system and the measurement results show that the server bandwidth saving rate of PPLive VoD system is as high as95%but that of UUSee VoD system is just70%. Also, the average number of interruptions during a viewing process in PPLive VoD system is as small as0.8, which means that the user viewing experience is wonderful. In a word, this strategy not only reduces server bandwidth but also improves the system QoS.更多还原