While streaming protocols like RTSP/RTP have continued to evolved, HTTP has remained a primary method for Web-based video retrieval. The ubiquity and simplicity of HTTP makes it a popular choice for many applications. However, HTTP was not designed for retrieving data with just-in-time tolerances; HTTP servers have always taken an as-fast-as-possible approach to data delivery. For media with known bandwidth constraints (e.g., audio/video files), HTTP servers can be enhanced and optimized by taking these constraints into account. For these data types, we present our architecture for an HTTP streaming server using paced output. We discuss the scalability advantages of our HTTP streaming server architecture and compare it with traditional HTTP server response times and bandwidth usage. We also introduce an intelligent bursting mechanism and consider its effects on end user experience.
Kevin J. Ma, Radim Bartos, Swapnil Bhatia