Edge Computing Aided Congestion Control using Neuro-Dynamic Programming in NDN

Named data networking (NDN) is an emerging network paradigm that decouples content from its storage location by providing one or more content copies and distributing them within the whole network. Congestion control is a fundamental and important problem in NDN, but it has not been well solved yet. Existing works can be divided into three main types, receiver driven flow based control, hop-by-hop interest shaping and hybrid control. While they are faced with more or less high computational complexity, multi-content source and multitransmission path problems, we proposed our edge computing aided congestion control scheme (EACC). The main idea is to detect congestion along the transmission path and avoid it by interest forwarding control at edge nodes. We add a new field to data packet to record the congestion status of the transmission path when it returns. After that, we deploy the core computing functions of the solution at edge nodes, and formulate the interest packet forwarding control into a local MDP (Markov Decision Process) problem based on the returned path congestion status and local user request information. Then we use neuro-dynamic programming (NDP) to solve this decision problem and present a practical implementation at edge nodes. The proposed scheme is implemented in ndnSIM simulator and compared to other two methods. Simulation results show the effectiveness of our scheme.