A hybrid model of the Akamai adaptive streaming control system

Video streaming is the application generating the largest fraction of the Internet traffic. Adaptive video streaming adds to classic video streaming the possibility of dynamically adapting the video bitrate to track the time-varying network available bandwidth, avoid playback interruptions and ensure the delivery of the best video quality. This work focuses on the adaptive video streaming control system employed by Akamai, a major Content Delivery Network operator whose video delivery system is used by several video streaming platforms, including Livestream. Differently from the typical client-side control, Akamai employs an interesting and unique hybrid client/server control architecture. In this paper we propose and experimentally validate a closed loop mathematical model of the control system in the form of a hybrid automaton. The model is analyzed to derive key properties which can be used to properly tune the controller parameters.