I. Introduction

Video currently accounts for more than 70% of the Internet traffic. It is projected that 82% of the Internet traffic will be made up of video in 2022, and live video streaming will contribute 17% of the Internet traffic [1]. To deliver a high level of user Quality-of-Experience (QoE), video needs to be streamed at high rate while avoiding video freeze and minimizing rate fluctuations. To achieve these goals in the face of dynamic network conditions, Video-on-Demand (VoD) streaming solutions, such as Dynamic Adaptive Streaming over HTTP (MPEG-DASH) [2] and HTTP Live Streaming (HLS) [3], prefetch video segments into a video buffer of 10 to 30 seconds or longer to maintain continuous and smooth video playback. For live streaming, users are additionally sensitive to the end-to-end (or screen-to-screen) video latency, namely the time lag from the moment when a video scene occurs till a user sees it on her screen. In the traditional TV broadcast system, the video latency with a mean of 6 seconds could be achieved [4]. By contrast, the current live streaming latencies on Over-the-Top (OTT) devices range from 10 to 30 seconds [5]. This long video latency could be detrimental for user QoE. For example, OTT users watching the World Cup Football final may have to wait for more than ten seconds to see a goal after their neighbors watching cable TV cheer for it. Long video latency will simply ruin the “live” experience.