基于用户预览行为的流媒体带宽分配研究

Research on Video Streaming Bandwidth Allocation for User Previewing Behaviors

流媒体用户通常有"预览"行为,对腾讯视频系统的测量结果显示即使在网络传输良好的情况下,约50%用户在离观看结束不到一半时就离开了.增加用户下载带宽在提高用户体验质量的同时也会带来服务器带宽浪费.本文主要贡献是:首先,建立了流媒体带宽分配的数学模型以刻画下载速率对带宽浪费率、单位时间服务用户数和QoE的影响;其次,发现增大下载速率,用户体验质量提升较大而带宽浪费较小(对腾讯视频系统的计算显示当下载速率是播放速率的两倍时带宽浪费率仅为25%),为适当增大下载速率提供了合理性依据;最后,给出了求解总用户体验质量最大化问题的最优下载速率的基本步骤.

A large number ot users exhibit ＂preview＇＂ benawors in vdeo streaming systems, By a measurement of me Tencent video system, about 50% users depart by the first half of viewing even for good network conditions. Although increasing downloading rate can increase QoE （ Quality of user Experience ）, it also waste bandwidth due to users＇ early departures. Our work makes the following contributions： First, we make a simple mathematical model to study the impact of downloading rate on bandwidth wastage, service rate and QoE; Second, we find that more downloading rate increases QoE more significantly while increases bandwidth wastage less significantly （ by the evaluation on the Tecent video system, only 25 % of the bandwidth is wasted when downloading rate is twice playback rate）, which justifies allocating more downloading bandwidth in practice. Third, we show how to solve the optimal down- loading rates to maximize the total QoE of users.