【作者】 苗艳超；

【导师】 孙凝晖；

【作者基本信息】 中国科学院研究生院（计算技术研究所） ， 计算机系统结构， 2004， 博士

【摘要】 随着计算机性能的提高和网络带宽的增加，以及视频压缩编码技术的进步，使得数字视频服务成为可能。本文主要讨论提供数字视频服务需要解决的几个关键技术，并实现一个原型系统。具体研究内容如下： 1 给出了一个基于通用PC服务器和非实时操作系统构建实时视频服务器的方法。通过对通用服务器进行软硬件扩展，使其能够输出符合DVB-C协议的视频流，针对有线电视传输网络，实现了一种MPEG-2传输流实时软复用算法和一种实时视频数据输出的流量控制策略。 2 给出了视频服务器服务质量保证的若干基本策略，包括：基于推拉结合的流量控制策略、基于丢包重传的差错控制策略、以及用于限制视频服务器负载量的接入控制策略。 3 基于机群结构的大规模视频服务器，提出了一种带宽控制策略，解决了机群内部数据调度通道与数据实时输出通道在存储系统上的带宽争用问题，从而满足输出数据的时限要求。另外，比较了多种内部负载调度和数据调度策略，提出了一种被动的第二级存储器命中优先法作为视频服务器的调度策略。 4 提出了一个视频服务器的高可用机制，即基于资源的多对多备份机制。一个主份资源由r个后备资源备份，同时一个后备资源也为多个主份资源提供备份，在任意不超过r个节点失效的情况下，失效节点的负载全部可由既定后备资源接管，并且接管后系统中没有节点过载。将资源的备份关系以r备份图(二部图)、资源图和失效图的形式表达出来，通过构造任意失效图都存在完全匹配的r备份图的方法，给出并证明了后备资源在任意分布和规则分布时，资源分配需要满足的充分条件，同时给出了一种在机群视频服务器中实现该机制的负载分配调度算法。更多还原

【Abstract】 It has become feasible to provide digital video service with improved performance of computers, enlarged bandwidth of networks, and progress in compressing technology of video data. In this dissertation, the relevant key technology is studied, and a prototype system is implemented. The results are as follows:1 An approach to establish real-time video server system with normal PC servers and unreal-time operation system is provided. The normal server is expanded in software and hardware to make output video streams satisfying DVB-C standard, and to provide video service by cable TV. The real-time soft multiplex of MPEG-2 transport stream and the flow control policies of real-time video data output are also implemented.2 The approaches to guarantee quality of service of video server are provided, including flow control, error control, and admission control policies of video server.3 The bandwidth control approach of the storage system is provided to deal with the competition between the internal data scheduling and the real-time data output in order to satisfy the deadline limitation of the data output. Furthermore, the load scheduling policy and cache disk scheduling policy are compared and the passive cache-hit priority approach is provided as the scheduling policy of the video server.4 A mechanism of high availability of video server is proposed. The mechanism is based on more to more resource backup. One resource has r backup resources and any reserve resource could backup for more other r esources. The load o f the failed nodes could be taken o ver by reserve resource and no overload arise in the system with the condition of no more than r failed nodes. The backup relationship is represented by r backup graph (bipartite graph), resource graph and failure graph. By making the complete matching r backup graph for the failure graph, the sufficient conditions of resource distribution are proposed and proved respectively with arbitrary distribution and regular distribution of reserve resource. The load distributing and scheduling algorithm to implement the mechanism in the cluster system is given.更多还原