Reinforcement learning-based algorithm for efficient and adaptive forwarding in named data networking

In order to solve the severe issues of current TCP/IP Internet architecture, Named Data Networking(NDN) recently attracts lots of researchers, deemed to be one of the most futuristic Internet paradigms among numerous Information-Centric Networking(ICN) proposals. Although the forwarding strategy is the key feature of NDN, it is still at a preliminary stage. Current forwarding strategies either base on the flooding strategy trying to reduce the side effect of Interest flooding or base on the route-driven strategy paying attention to decrease the extra cost in maintaining the routing information. They are both deficient because of the storm issues or too much extra overhead. In this paper, we present the feasibility of using reinforcement learning algorithm such as Q-Learning in NDN. By modifying Q-Learning algorithm to solve the inherent issues, we design and implement IQ-Learning(Interest Q-Learning) strategy and DQ-Learning(Data Q-Learning) strategy, which learn from the past experience and make the best forwarding choice. The simulation results on NDNSim show that our forwarding strategies achieve higher Interest satisfaction ratio, shorter Interest satisfaction delay and are more sensitive to the changes of network than Flooding strategy and the state-of-the-art BestRoute strategy.