NDN Producer Mobility Management Based on Echo State Network: A Lightweight Machine Learning Approach

NDN is one of promising underlying network architectures that supporting 5G because of its characteristics such as decoupling senders and receivers, hop by hop transmission, in-network caching, etc. However, it still faces challenges in producer mobility management like the triangle routing problem (non-optimal routing path) and global centralization of the home agent, causing a poor scalability in large network scales and long handover delay. In this paper, we propose the ESN - PBA, a NDN producer mobility management scheme using the ESN in prediction to realize a lightweight machine learning based seamless handover algorithm. Better than the existing fixed and post-adjustment management schemes of producer mobility management, the ESN-PBA can perceive nodes movements heuristically and pre-configure the adjustment in advance to reduce overall processing overhead. In addition, with fine-grained home router status feedbacks and NDN content data oriented philosophy, the training process of normal machine learning method can be mutually enhanced. The experimental results in ndnSIM show that, in the case of successful prediction, the effect of seamless handover can be achieved straightly on the fly. In order to improve the hit rate of cache, we take advantage of NDN's multipath forwarding support, the utilization of ESN prediction of multiple candidates and synchronous forwarding. Compared with PIT-based approach and DNS-based approach, the handover delay of ESN-PBA reduces by 66.7% and 75% respectively. Besides, its handover overhead reduces by 38.4 %, compared with DNS-based approach.