基于http streaming的课程直播系统中流传输子系统的优化-optimization of streaming subsystem in curriculum live broadcast system based on http streaming.docxVIP

基于 HTTP Streaming 的课程直播系统中流传输子系统的优化摘 要HTTP Streaming 技术的主要思想是将视频流分割成多个小的分片 文件并且存储于 WEB 服务器，客户端通过 HTTP 协议下载需要的分 片文件进行播放。由于 HTTP Streaming 具有很多优点，例如，该技术 基于广泛使用的 HTTP 协议，可靠且易于部署；对于防火墙或 NAT 的 穿透性好；可以利用标准的 HTTP 服务器以及 HTTP 缓存等，该技术 已经成为近年来流媒体传输领域的热点。本文工作围绕将该技术实际 应用在大规模课程直播系统中时暴露出的若干不足展开。HTTP 缓存服务器是提高 HTTP Streaming 系统客户并发量的关键 环节。但当前主流 HTTP 缓存服务器，如 Nginx、Squid、Varnish 等， 在缓存资源更新期间的行为都存在不足，在我们实际部署的基于 HTTP Streaming 课程直播系统中，缓存服务器会周期性地把大量客户端请求 转发至源服务器，从而制约了 HTTP Streaming 系统的可伸缩性。本文 针对性的提出一种优化的 HTTP 缓存服务器在缓存更新期间的行为， 即缓存服务器仅向源服务器转发一路客户端请求，缓存更新期间，拒 绝其它关于该资源的请求。优化策略在使用最为广泛的 Nginx 服务器 的基础上进行了实现。实验证明，优化后系统的伸缩性得到了显著提 高。另外，本文还对主流 CDN 服务的缓存行为进行了检测。典型的 HTTP Streaming 直播系统的端到端延时在 10s-15s 之间， 这对于需要支持一定交互性的课程直播应用来说延时过大。通过对引 入端到端延时的各个环节（客户端的包调度机制、客户端去抖动缓冲、 分片文件长度）的优化，本文提出一套低端到端延时的 HTTP Streaming的直播原型，成功将直播系统的端到端延时降低至 5s 以下。关键词：HTTP Streaming, 缓存服务器， 缓存更新， 端到端延时OPTIMIZATION ON STREAM TRANSMISSION SUBSYSTEM OF HTTP STREAMING BASED COURSE LIVE BROADCAST SYSTEMABSTRACTHTTP Streaming is the technology which cuts the media streaming into fragments and puts them on web server, while clients download such fragments through HTTP and play them sequentially. HTTP Streaming has lots of advantages, for example, based on the widely used HTTP, it is stable and reliable; it easily avoids NAT and firewall traversal issues; it enables the ability to utilize standard HTTP servers and HTTP caches. Such technology has been quite popular in the field of media streaming. This paper mainly focuses on fixing the shortages those appear when HTTP Streaming is applied in the large-scale course live broadcast system.HTTP Cache Server plays a key role in increasing the scalability of HTTP Streaming systems. However mainstream cache servers, such as Nginx, Squid and Varnish, behave improperly when the cache file is being updated. When used in live HTTP Streaming systems, this behavior will cause the cache server to deliver many unnecessary requests to the origin server periodically so that the scalability of the system