Snowland:Fair,Efficient And Stable Multi-server HTTP Adaptive Video Streaming

We have witnessed for years that HTTP adaptive video streaming,where a player can dynamically adjust the video bitrate in response to changing network conditions,has been widely adopted in industry.On the downside,when multiple players compete for a bottleneck link,it often suffers from lackluster performance with respect to three important properties: fairness,efficiency,and stability.To tackle these problems,effective algorithms have been proposed in the case of singleserver streaming where each player requests video only from a single web server.Due to the insatiable appetite for bandwidth when playing high-definition video streams with high bitrates,industrial companies have extended singleserver streaming to multi-server streaming,where a player requests video chunks from multiple web servers concurrently.Thus,we revisit the three properties in the context of multi-server HTTP adaptive streaming,and show that these properties no longer hold with existing algorithms designed for the single-server case.In this paper,we present a new protocol designed to offer the best possible user experience by offering fair,efficient and stable multi-server HTTP adaptive video streaming.Specifically,in this paper,a bitrate adaptation algorithm is proposed to achieve fairness across players when players employ multiple flows.Furthermore,to improve efficiency and stability,we present a flow scheduling algorithm which adjusts the number of flows based on current network conditions.Our extensive real-world experiments demonstrate that our protocol achieves fairness,efficiency and stability in the context of multi-server streaming,and substantially outperforms its closest alternative.