Abstract:
Dynamic Adaptive Streaming över HTTP (DASH) became the pillax of mul timedia content delivery mechanisms in the last decade. Given fluctuating network conditions, over-the-top content platforms struggle with delivering a high quality of ex perience (QoE) and uninterrupted playback sessions. To overcome this difficulty, they keep multiple quality levels of the same content in a fragmented way. Tlıis mechanism enables players to adapt the video quality to varying network conditions by changing the video bitrate at the fragment boundaries. To maxinıize the QoE, adaptive bitrate algorithms have been widely studied in the literatüre with promising results. However, the majority of the state-of-the-art solutions do not take into account the presence of multiple DASH clients on the shared bottleneck link, whereas the existing studies con sidering multiple DASH players in the same network do not consider the diversity of fragment durations among different video titles, background traffic and users' privacy. Those gaps cause QoE fairness and stability problems along with feasibility concerns. First, to address these problems, we propose a centralized modüle assisted adapta tioıı mechanism with a lightweight Software-Defined Networking (SDN) integration for on-demand video streaming. Second, we leverage our proposed SDN-assisted mech anism to deliver QoE-driven low-latency live event streaming över HTTP. Third, we implement a live streaming DASH client with a novel bandwidth measurement heuris tic without requiring any extra component to the legacy systems. It reduces the live delay betvveen the actual event to users' screens down to İs. As a final contribution, we present a deep reinforcement learning framework to adapt the playback speed and video bitrate to maximize QoE in live streaming
