【作者】 阴进涛；

【导师】 金海；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2007， 硕士

【摘要】 随着宽带的普及和对等网络技术的发展,现在大部分流行的网络直播平台为了解决传统客户机/服务器模式的网络直播扩展性差、系统成本高等问题都采用对等计算技术交换流媒体数据。现有对等网络直播平台大都构建树状或简单网状网络逻辑拓扑。树状结构在现有互联网络状况下扩展性和维护性差,难以保证低延迟,特别是上层关键节点的失效会对整个网络逻辑拓扑产生非常大的影响,因此越来越多的系统开始基于网状拓扑构造,网状拓扑管理机制的研究也越来越重要。对等网络直播平台AnySee(版本4.0,除非特指,下文简称AnySee)中的网状拓扑管理机制采用优化的邻近节点选取算法和成员节点管理算法构造网状逻辑拓扑,增强了节点的空间邻近性,从而便于系统的扩展和维护;采用局域网选代表算法使得网络逻辑拓扑与底层物理拓扑相适应,减少了网络逻辑拓扑中不必要的带宽瓶颈,增加了系统传输流媒体数据的效率。系统通过分布式索引服务器获取初始的成员节点信息,增强了节点的地域邻近性,加快了初始成员节点的获取速度,减小了平台的启动时间;索引服务器辅助客户端构造网络拓扑,邻居节点间通过相互通信维护网络拓扑,使得节点仅仅依靠局部信息即可获得空间和时间邻近的邻居节点,从而降低了数据延迟,提高了伙伴节点的服务质量。通过上述机制的实施,使得平台获得了更好的扩展性和稳定性。对等网络直播平台AnySee基于Windows操作系统和Visual Studio平台,采用C++语言实现。AnySee平台中的网状拓扑管理机制采用优化的网络拓扑构造算法和拓扑管理技术,使得平台在扩展性,维护性,稳定性和可靠性方面均有提高。对邻居节点选择成功率、缓冲区满度概率/播放连续性、流媒体服务器负载、索引服务器的压力的测试表明网状拓扑管理机制使AnySee平台的媒体服务器的CPU占用率降低了50%,内存占用率降低了30%,网络带宽占用率降低了45%,有效地提升了AnySee平台的性能。更多还原

【Abstract】 With the prevalence of broadband and progress of peer-to-peer technology, most of nowadays prevailing network live streaming platforms are all based on Peer-to-peer technology, solving the poor-scalability and high-cost problems of traditional Client/Server-based live streaming systems. However, the existing Peer-to-peer -based live streaming platforms are builded upon tree-based or simple mesh-based overlay network. Tree-based structures are hard to scale and maintain, no guarantee of low latency, especially upper key node’s failure leads to fatal breakage of entire overlay network. As a result, more and more systems turn to mesh-based overlay construction.Mesh-based overlay construction strategy for peer-to-peer live streaming platform AnySee (Version 4.0, AnySee for short below unless mentioned) using optimized adjacent node choosing algorithm and leaguer management to construct mesh-based overlay, improves the scalability and maintenance of the platform and using LAN representative algorithm to fit overlay to underlying physical topology, reduces unnessary bandwidth bottlenecks of overlay and improves the efficiency of transmission of streaming media data. Platform using distributed tracker server to retrieve initial adjacent nodes of node and record all of nodes’information, greatly accelerates the speed of initial member node selection and reduces the startup time of platform. By having tracker system take part in overlay construction and communication among nodes maintain overlay, with local information, node gets space and time near member nodes, which reduces data latency and enhances QoS of partner nodes. Platform is more scalable and stable.Peer-to-peer live streaming platform AnySee is based on Windows and Visual Studio and using C++ programming language. Mesh-based overlay construction strategy for AnySee uses optimized overlay construction algorithm and overlay management technology, so that the platform scalability, maintenance, stability and reliability have been improved. Measurements of successful rate of member node selection, buffer full probability/playback continuity, load of broadcast server and pressure of tracker server suggest that new mesh-based overlay construction strategy improves platform performances of AnySee, compared to Client/Server systems, Source server’s CPU usage is decreased by 50%, memory usage by 30% and bandwidth by 45%.更多还原