Liv(e)-ing on the Edge: User-Uploaded Live Streams Driven by "First-Mile" Edge Decisions
Jiasi Chen, Bharath Balasubramanian, Zhe Huang · 2019
As users equipped with high-resolution cameras spontaneously capture and live-stream videos to interested parties through distributors like Twitch or Facebook Live, a new style of video delivery is emerging: user-upload streaming. Unlike video-on-demand and traditional live-streaming in which video is stored by professional content providers and distributed using standard CDN techniques, in user-upload streaming, the video is generated and uploaded on-the-fly by heterogeneous clients with restricted bandwidth and long latencies to the distributor. Hence, it is crucial to factor in client-side "first-mile" factors when making edge decisions. In this paper, we present a systematic design of a video delivery architecture for user-upload streaming, which focuses on the upload server and upload bitrate as the important "first-mile" edge factors that influence downstream delivery and quality-of-experience for viewers. We present a polynomial-time algorithm to minimize the end-to-end latency and maximize the video rate for all users for the scenario where the upload and download server are the same. Further, we present efficient heuristics for the general (NP-complete) version where the upload streams are routed through the distributor's overlay network. Finally, we validate the efficacy of our algorithms through extensive trace-based simulations based on real-world data sets.