Figure 1. Open-Loop Rate Control Architecture (OLAC)

The video has been an integral part of communications and entertainment applications for many years. Due to the growth and popularity of the Web, video delivery via best-effort packet networks such as the Internet has become an important research area. Nowadays, video streaming has become a dominant traffic type for the Internet and its usage will constantly increase in the near future. To be effective, videos need to be presented to users in a high quality and without any interruptions. However, video delivery via best-effort packet network faces a number of challenges, including unknown and time-varying bandwidth, delay and losses, and scalability issues.

Today, adaptive streaming is the prevalent technology for supporting video streaming over TCP by offering multiple quality levels for a video stream that is segmented into chunks. Currently, it is a standard technique for Internet video streaming over HTTP (dynamic adaptive streaming over HTTP, DASH). Extensive modelling and evaluations of adaptation schemes have revealed that state-of-the-art adaptive solutions only reach about 50% of the achievable throughput under buffering of several deciseconds. To a large extend, this performance bottleneck is a result of the underlying TCP transport layer and the biased rate control.

In contrast to existing work, our project focuses on the research of a low-latency video delivery service with adaptive solution that fulfills the requirements of live video broadcasting. Our previous works [Shuai et al. CCNC’15, Shuai et al. ICC’15, Shuai et al. CCNC’16] show that a server-based rate control architecture (Figure 1) using client buffer simulation provides low-delay feedback. At the same time, hybrid rate adaptation logic that is based on both the throughput and buffer information stabilizes the adaptive response to buffer dynamics. Altogether, it allows video streaming with very small buffer sizes, in fact as small as a video-chunk duration, and it achieves significant improvement of state-of-the-art solutions with respect to user-perceived quality. Furthermore, we have built an OpenFlow-based testbed at Saarland University and have developed a web-based visualization application for the testbed, which enables easy deployment of innovative network concepts and extensive network analysis in a real wide-area network.

Our contributions in the project will address specific challenges in previous work required for low-latency streaming at a high quality. In particular, we will design and develop a new system for adaptive streaming that minimizes the required video buffers on clients and scales in large deployments for broadcast events like sports and concerts. Besides extensive evaluations, we will also validate the performance of the new system in the Software-Defined Networking (SDN) scenario. An OpenFlow-assisted adaptive streaming shall optimize the user Quality-of-Experience (QoE) for all streaming devices in a network, whilst various device and network requirements are taking into consideration.

Project Team

Principal Investigators
Prof. Dr.-Ing. Thorsten Herfet

Co-Principal Investigator
M.Sc. Yongtao Shuai