A Reinforcement Learning Framework for Multi-source Adaptive Streaming

Dynamic adaptive streaming over HTTP (DASH) is widely used in video streaming recently. With DASH, a video is stored in multiple equal-playing-time chunks with different quality levels. Video chunks are in-order delivered from a single source over a path in traditional DASH. The adaptation function in video player chooses a suitable quality level to request depending on current network status for each video chunk. In modern networks such as content delivery networks, edge caching, content-centric networks, etc., popular video contents are replicated at multiple cache nodes. Utilizing multiple sources for video streaming is investigated in this paper. We propose a reinforcement learning based algorithm, called RAMS, for rate adaptation in multi-source video streaming. The proposed algorithm outperforms the other notable adaptation methods.