Congestion-aware quality-adaptive streaming of scalable video

Abstract

Internet video streaming is a hot topic in multimedia systems. A large variety of devices (computers, mobile phones, TVs, etc.) are connected to the Internet via wired or wireless networks and are capable of receiving and decoding HD video content. To enable new services like HD video streaming (e.g., online video rental), the Internet’s infrastructure was enhanced. But the Internet is still a best-effort network, which does not implement quality-of-service or admission control, resulting in time-varying bandwidth and packet delay, packet loss and network congestion. Because video streaming accounts for a considerable amount of the Internet’s traffic, video streaming needs additionally to be congestion-aware, to avoid a congestion collapse of the Internet. The Transmission Control Protocol (TCP) can adapt to changing network conditions and is currently the de facto standard protocol for congestion-aware and reliable data transmission in the Internet. This fact gave TCP-based video streaming a huge momentum. Consequently, this thesis investigates TCP-based adaptive video streaming for the Internet. The main goal is to provide a solution for congestion-aware video streaming, while still being able to achieve a reasonable performance in error-prone networks. To complement existing work on congestion-aware adaptive streaming, this thesis makes six contributions. (1) The baseline performance of TCP-based adaptive streaming is identified by means of an evaluation of different adaptive streaming approaches. The results represent a reference for further investigations. (2) An investigation on the influence of TCP’s behavior in presence of packet loss on the video streaming performance. (3) To overcome the shortcomings of TCP-based video streaming (single TCP connections fail to deliver a good performance in case of packet loss), a new approach to video streaming based on multiple request-response streams was introduced. The novelty of this system is that it is able to make use of multiple HTTP-based request-response streams while still providing TCP-friendliness. (4) A performance model of the HTTP-based request-response streams was developed, to estimate the influence of the system parameters and the network characteristics on the throughput performance. (5) A comprehensive evaluation of the HTTP-based request-response streams under diverse network conditions was conducted, to validate the model’s estimations. Additionally, the TCP-friendliness was evaluated, showing that request-response streaming systems can be configured to achieve TCP-friendliness. (6) A cellular network with high bandwidth fluctuations and RTTs was used to investigate the performance of the request-response streaming system in a mobile video streaming scenario. The results indicate that the streaming system can make good use of the available bandwidth, while the number of quality switches is kept low. While aggregating multiple TCP connections to improve the TCP streaming performance is quite common, usually the improvement comes at the cost of high deployment effort. By placing the streaming logic at the client, request-response streams can avoid this complexity. Additionally, this client-driven approach responds faster to changing network conditions and enables easy recovery from connection stalls or aborts, because the control loop is at the client. To improve the network efficiency and the scalability in terms of number of clients served, HTTP-based request-response streams can utilize HTTP proxies and caches.