Dynamic Adaptive Streaming Control based on Deep Reinforcement Learning in Named Data Networking

Named Data Networking (NDN) is a general proposed network layer protocol which offers a set of rich functionality: in-network storage, multi-path forwarding, and data-centric security. The cache and multi-path feature improves the efficiency of transmission but increases the difficulty of the bandwidth estimation.In dynamic adaptive streaming,video of different quality is segmented in server,clients could select the most appropriate segment to download due to the network status. In this paper, a deep reinforcement learning method is proposed for dynamic adaptive video streaming over NDN to maximum user-perceived qualityof-experience(QoE).As a model-free method,our algorithm doesn't need accurate bandwidth estimation. It can use all kinds of information during the video playback process to make a sensitive decision. Experimental results indicate that our algorithm performs better than others in NDN.