Adaptive scalable Internet streaming

This thesis presents a comprehensive investigation of the performance of real-time streaming in the best-effort Internet and studies several novel congestion control and retransmission methods of real-time content delivery over the public Internet. As our experiments show, these new methods provide a significantly better quality-of-service (QoS) to the end-user than the existing methods.; The first half of this work focuses on constant-bitrate streaming over the existing Internet in a large-scale performance study and examines both the behavior of network parameters and the quality of video experienced by the end users. We model select network parameters and gain an insight into what conditions future streaming applications should expect in the best-effort Internet. Furthermore, we extensively study the performance of real-time retransmission (in which the lost packets must be recovered before their decoding deadlines) based on our traces and present new methods of reducing the amount of duplicate packets generated by the server in response to lost packets.; The second half of this thesis studies congestion-adaptive streaming in the best-effort Internet. Using scalable MPEG-4 layered coding as the target application, we develop new congestion control methods that are used to rescale the enhancement layer to match the available bandwidth in the network. We find that traditional ACK-based congestion control used in TCP is not flexible enough to be applied to real-time streaming. On the other hand, as our work shows, rate-based (or NACK-based) congestion control typically does not scale to a large number of flows. To overcome this difficulty, we present a novel rate-based congestion control scheme that scales well while satisfying all requirements of a real-time application. To support this scalable congestion control, we find that the flows must possess the knowledge of the bottleneck bandwidth of an end-to-end path. Consequently, as part of this work, we present an extensive performance study of bandwidth estimation methods that can be used in real-time by the client to supplement its rate-based congestion control with the value of the bottleneck capacity.