基于分支过程理论的半结构化覆盖实时媒体组播延迟性能建模与分析

项目名称： 基于分支过程理论的半结构化覆盖实时媒体组播延迟性能建模与分析

项目编号： No.60872053

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 王晖

作者单位： 中国人民解放军国防科学技术大学

项目金额： 24万元

中文摘要： 半结构化覆盖实时媒体组播（Semi-sTructured Overlay Real-time media Multicast，STORM）包括流媒体直播与流媒体点播。延迟性能是流媒体直播系统设计考虑的关键因素之一。项目组运用依龄分支过程对对等网络数据组播延迟性能进行建模，分析了其随节点规模与增长速率的演化规律，得到了多个数据块同时传输情况下组播传输延迟的下界。为了验证延迟性能模型及其分析结果的有效性，项目组设计了一个多协议P2P流媒体爬行引擎TVCrawler，对多个P2P TV系统中进行了面向仿真的用户行为模型研究，从而获得用户行为规律、会话时长分布、用户到达率以及网络动态性等仿真验证实验设置中的关键参数，并设计实现了基于数据包级的多线程仿真验证平台。服务器负载性能是影响流媒体点播系统延迟性能的主要因素。项目组运用基于微分方程的流模型理论等对对等网络流媒体点播系统的服务器负载性能进行了建模分析，提出了节点相互贡献率模型以及基于概率的服务器负载计算模型，并且引入主动协同网络以及网络编码对服务器负载性能进行优化。

中文关键词： 半结构化实时媒体组播；延迟性能；分支过程；服务器负载

英文摘要： Semi-sTructured Overlay Real-time media Multicast (STORM) possesses two kinds of commonly adopted applications: live streaming and video-on-demand (VoD). Delay is one of most important performance metrics that should be taken into account when designing a live streaming system. We extend the Snow ball algorithm based on Age-dependent branching process. This extended algorithm enables us to obtain lower bound of delay for multiple data blocks transimiting at the same time, given peer population. We propose a delay model using Age-dependent branching process and analyze the relationship between delay performance and peer population. We also design a multi-protocol crawler to perform a simulation-oriented study on user behavior models in multiple real-world commercial live streaming systems, including session length, arrival rate and dynamics, etc. These results enable us to verify the delay model and analytical result in a more practical setting with the packet-level multi-thread simulation system we design. For P2P-VoD systems, server loading is a key factor which influences the delay performance. We propose models to analyze the server loading in P2P-VoD systems by implementing stream theory based on differential equation. We propose a peer contribution model and a computational model of server loading based on probability theory. Helper network and network coding are introduced into optimizing server loading.

英文关键词： STORM; Delay performance; Branching process; Server loading