Research on Adaptive Congestion Control techniques for Internet Video Streaming

Video streaming has strict bandwidth, delay and packet loss requirements, but the best effort services of current Internet can not offer any quality of service (QoS) guarantees to real-time traffic. Congestion control is required to reduce bursty loss and excessive delay, but packet loss is still unavoidable, thus error recovery mechanism must be in place to maximize video presentation quality in presence of packet loss. Most of recent congestion control and error recovery researches are separated, however, the combination of these two mechanisms would improve the performance of Internet video streaming. In this paper, a layered end-to-end unicast Internet video streaming architecture is presented. Specifically, an RS coding scheme is designed in error recovery layer against packet loss. A flexible congestion control scheme, AVSP (Adaptive Video Streaming Protocol) is designed in video transmission layer against bandwidth fluctuation. There are four states in AVSP, including: paused, quick\|start, steady and congested. AVSP applies a weighted moving average rate adjust policy by examining whether the packet loss rate exceeds the threshold value or not. The alterable parameter setting of AVSP enables that it could be used under variant network conditions. Based on the Gilbert model and the serial number of packets, AVSP could get an estimated packet loss rate for next rate adjust decision. Its adaptation capacity could be compared with the Additive Increase and Multiplicative Decrease (AIMD) algorithm used in TCP in the same bandwidth fluctuation condition. Considering that the Internet is a shared environment, the inter-protocol characteristic (especially the TCP-friendliness) should be taken into consideration in AVSP design. The TCP-friendliness could be measured by an average throughput comparison in a simulation case with several concurrent AVSP flows and TCP flows. Simulation results demonstrate that AVSP has a good adaptation capacity against available bandwidth fluctuation, and act with TCP-friendliness in a shared environment.