A SMDP-based forwarding scheme in named data networking

Named data networking (NDN) provides a promising networking paradigm for efficient content delivery. In NDN, adaptive request forwarding is inherently supported by enabling routers to dynamically select the next hop for each request based on the network conditions. However, due to the limitation in network resources, a well-designed forwarding strategy is necessary to achieve satisfactory network performance. In this paper, the problem of content request forwarding in the context of NDN is naturally formulated as a semi-Markov decision process (SMDP). Since the exact SMDP solution is intractable, a variant of reinforcement learning (RL) method integrated with function approximation by neural networks is adopted to find the optimal solution for our SMDP abstract. A broad set of experimental comparisons was carried out to verify the effectiveness of the resulting forwarding strategy. The simulation results show that the RL method provides an efficient solution to our SMDP-based forwarding model and our approach can further enhance network performance in comparison to existing forwarding strategies by reducing rejection ratio, network load, as well as delivery time. Load balance and differentiated services are also considered in our proposal.