Viewer-Oriented CDN Scheduling on Crowdsourced Live Video Stream
Wen I. Yang, Yuhan Hu, Lan Ding, Ye Tian · 2019
In crowdsourced live video stream platforms, viewers are spread all over the world. To improve the Quality of Experience (QoE), streams are distributed to the content distribution network (CDN) deployed in different regions, and viewers are assigned to the closest CDN to watch live broadcast. Currently, in these platforms, like Douyu and Twitch, the stream distribution strategy is based on the popularity of stream, without considering viewers' engagement, which is significant to the platform. In this paper, taking Douyu as a case, we conduct a detailed analysis of the platform. Our results show the gap between popularity based stream distribution strategy and platform's profit. To fill the gap, we propose a viewer-oriented algorithm to distribute stream and determine viewer's streaming source based on viewer's engagement, which targets maximize platform's profit. We examine our algorithm using real trace of Douyu. The results indicate that our algorithm can improve the overall profit of platform and uniformity of load in CDN with low scheduling cost.