Integrating User Behavior in Interactive Streaming Bitrate Throttling

Abstract

In interactive video streaming, compression algorithms face a tradeoff between image quality, as defined in terms of brightness and color accuracy, and com- pression size. A heavily compressed video stream will take up less bandwidth, suffer less packet drops and be faster to encode and decode, but will come with terrible image quality. At the same time, video data keep risings as the largest amount of data transferred over the Internet; a situation which is bound to accelerate as interactive video streaming through cloud gaming, remote desktop and application streaming become more and more prevalent. Many researchers have devised different definitions of Quality-of-Experience (QoE) in Internet video streaming. I present an alternate definition to QoE applied to interactive Internet video streaming that factors in the user’s actions in the form frames- per-second and frame pixel differential sizes, and show that can be combined with existing QoE metrics like buffering and packet drops-based approaches. I then present a new bitrate throttling algorithm which incorporates the user’s behaviors on top of network conditions via a dissatisfaction score based upon the QoE metric I define, and apply it experimentally to a remote desktop scenario. An experimental comparison of a remote desktop situation with different workloads having different and varying frames-per-second and image quality requirements shows that user-behavior augmented bitrate throttling is a more accurate throttling algorithm at delivering a better and ore consistent interactive streaming experience over the Internet, under both constant and fluctuating network conditions.