Network-congestion-aware video streaming: A rest-and-download approach

摘要

On-demand video services such as Youtube and Hulu are expected to comprise a large percentage of the increasing data loads in mobile networks. On-demand video is distinctive because it is pre-recorded and therefore can be considered elastic traffic because the video frame buffer can be downloaded well past the current point of playback. Based on this observation, we propose Video Rest-and-Download (VR&D) as a video download application framework that aims to reduce network congestion while maintaining playback quality. The intuition for VR&D is that, in a scenario where radio resources are shared by multiple data users, the video user can “rest” for some amount of time until fewer users are in the network, thereby allowing other data users to complete their downloads faster, without affecting playback quality. We present an algorithmic framework for VR&D based on the Markov Decision Process that uses the history and current state of network activity to determine how aggressive the user should be in downloading video frames. We evaluate its performance using a simulated UMTS network with HSDPA data service based on real network traces from a major U.S. carrier. Our results show that, compared to the existing solution, during the time of video playback this application can reduce download time by as high as 50%, and alleviate network congestion by up to 30% with minimal effect on playback quality.