【作者】 王浩；

【导师】 钟玉琢； 杨士强；

【作者基本信息】 清华大学 ， 计算机应用技术， 2002， 博士

【摘要】 Internet视频组播是众多的网络多媒体通信应用中的重要组成部分，具有广阔的应用前景。但是由于当前的Internet无法对视频流传输提供服务质量的保证，另外由于Internet和用户的异构性，因此Internet视频组播研究面临着巨大的挑战，很多问题需要更进一步的研究。本文结合新出现的视频压缩编码技术以及Internet网络传输协议的新变化，重点研究了视频组播传输领域中的三个子课题：1、具有细粒度码率调整能力的视频组播算法；2、利用Peer-to-Peer技术增强应用层视频组播的性能；3、基于视频组播技术的流合并调度策略。本文的创新和研究成果如下：1. 提出了一种具有细粒度码率调整能力的动态分层视频组播体系结构－FGAVM现有的视频组播算法中，多视频流重复组播算法网络动态适应能力强，但是带宽利用效率低；分层视频组播算法提高了带宽利用效率，但网络动态适应能力差。为了提高视频组播的性能，本文结合细粒度可扩展视频编码技术，提出了一种具有细粒度码率调整能力的动态分层视频组播体系结构－FGAVM。FGAVM充分利用新的视频编码技术的特点，采用分布式和集中式相结合的码率控制策略，网络带宽利用效率高，码率调整粒度精细，兼顾了现有的视频组播算法的优点。实验结果表明：FGAVM在适应网络异构性、适应网络动态变动的能力方面均优于现有的其它算法，同时它还具有多视频会话可公平共享带宽、具备良好的TCP友好性等优点，是一种更为理想Internet视频组播体系结构。2. 提出了一种基于Peer-to-Peer技术的应用层视频组播传输结构－P3S受当前备受关注的Peer-to-Peer文件共享服务的启发，本文对现有的应用层视频组播树型传输结构进行了扩展，提出了一种新的网格状层次型传输结构－P3S。本文重点研究了P3S结构中的两种Peer-to-Peer协作技术－错误恢复技术和内容分配技术。其中，错误恢复技术利用用户之间的数据交<WP=8>换以降低最终的数据丢失率。为了消除数据包丢失相关性造成错误恢复技术的性能下降，本文通过数学建模、理论推导和实验仿真，证明了通过引入一个小的传输延迟，可以使得该技术在各种数据包丢失模式下均取得良好的效果；内容分配技术是指视频发送（转发）节点根据下游用户的网络带宽状况对传输内容进行划分，为每个用户提供不同的数据，通过下游用户之间的数据交换，可以获得更大的全局数据吞吐量。本文提出了一种基于线性规划的全局最优内容分配算法，同时，为了降低算法的复杂性，提出了一种单步最优的启发式“贪婪”算法。实验结果证明，与现有的应用层视频组播方法相比较，结合了Peer-to-Peer用户协作技术的P3S视频组播方法的最终数据丢失率低、全局数据吞吐量大、视频图像质量高，明显提高了视频组播质量。3. 提出了一种基于视频组播技术的视频流合并调度策略－Patching FirstPatching First调度策略以现有的Patching流调度算法为基础，利用视频组播技术，将点播同一节目的用户请求合并服务，提高系统吞吐量。为了提高原有算法的性能，Patching First调度策略区分系统中的Patching流和Regular流的重要性，提出应该优先调度资源占用更少的Patching流。根据Patching First流调度策略，本文提出了一种基本的等待队列调度算法MPQL，在MPQL的基础上，结合批处理算法的思想，提出了对MPQL的改进算法MFPQ。实验结果证明，与原有算法相比，本文所提出的基于Patching First流调度策略的两种等待队列调度算法可以有效降低用户点播请求的平均撤销率、请求撤销的不公平性以及用户平均等待时间。更多还原

【Abstract】 Multicasting video over the Internet is an important component for various modern multimedia communicating applications. Although video multicasting shows a promising future, it has to overcome two main obstacles, 1. Current Internet infrastructure can not guarantee the Qualility of Service for video streaming, and 2. The networks and users connected to them show enormous heterogeneity. Many problems are still remained in this area and more research work is expected.Combining with new video coding techniques and emerging network protocols, this paper focuses on three research topics in multicasting video over the Internet. These topics are: 1. Fine-grained rate adaptive video multicasting algorithms, 2. Using peer-to-peer techniques to enhance the performance of application-level video multicasting, and 3. Scheduling strategies based on stream-merging using video multicasting techniques.The creativities and contributions are discussed in detail as follows：1. This paper proposes a fine-grained rate adaptive dynamic-layering video multicasting framework - FGAVM. Among current video multicasting algorithms, Replicated-Stream Multicast adapts the rate well according to network dynamics, but its bandwidth efficiency is poor; On the other hand, Layered Video Multicast has high bandwidth efficiency but its adaption is poor. To bring all the merits of existing alortithms together while avoiding their drawbacks, this paper harnesses the power of FGS video coding technique to propose a new video multicasting framework - FGAVM. FGAVM adopts a hybrid of centralized and de-centralized strategies to adjust the bitrate of each multicast channel according to network dynamics with fine granularity. Extensive simulation results show that FGAVM can fine-tune the bitrate and also achieve high bandwidth efficiency. Moreover, FGAVM can provide inter-session fairness and TCP friendliness thus is a better framework for Internet video multicasting.2. Using peer-to-peer techniques, this paper proposes a new application-level video multicasting architecuture called P3S,. Inspired by the emerging peer-to-peer file sharing service on the Internet, this paper extends the delivery tree structure of current application-level video multicasting and proposes a grid-based hierarchical architecture - P3S. In P3S, service subscribers can cooperate with each other by using peer-to-peer techniques. This paper concentrates on two cooperative techniques between service subscribers in P3S: Error Recovery and Data Allocation. Error recovery technique aims at decreasing the overall packet loss ratio by interchanging<WP=10>data between subscribers. In order to combat the impact of correlative packet losses, this paper proposes to introduce some controlled delay between two unicast sessions originated from the same intermediate P3S node. Both theoretical and experimental results support the proposed technique. Data Allocation means that the source node of each P3S grid allocates different part of the the video content for each downstream receivers according to the bandwidth between them. After the cooperation between the two receivers, each receiver can achieve higher data thoughput. In this paper, we propose a technique, based on linear programming that optimally allocates the content so that the total throughput of the whole session is maximized. We also present a greedy data allocation algorithm that is slightly suboptimal but much simpler. The simulation results show that, compared with current application-level video multicasting methods, P3S with the techniques we proposed has the advantages of lower packet loss ratio, higher overall data throughput and higher quality of video image, thus can provide better performance for Internet video multicasting.3.Based on video multicasting techniques, this paper proposes a new scheduling strategy called Patching First. Patching First improves the performance of current Patching algorithm by serving user requests differently according to their different types. By allocating resources for low-co更多还原