A Q-learning based Forwarding Strategy for Named Data Networking

Interest packets forwarding is one of the key features of Named Data Networking (NDN). Traditional Interest forwarding strategies exploit the faces setup by the routing protocol in the Forwarding Information Base (FIB) table. The main drawback of such strategies is that the changing network conditions can invalidate the forwarding rules in the FIB. Recent strategies do not rely on the routing information and actively seek new delivery paths using reinforcement learning techniques. However, these solutions usually introduce modifications in the NDN data structures and/or packets. In this paper, we devise a new lightweight forwarding strategy, dubbed QRF, using the Q-learning algorithm. QRF does not rely on the FIB table while preserving all NDN-related data structures and packets. QRF performs online learning to adapt its forwarding decision according to the state of the Pending Information Table (PIT). Simulation results show that QRF outperforms BestRoute in terms of delivery time in congested scenarios.