A Hybrid Strategy for Addressing Interest Flooding Attacks in Named Data Networking

Named Data Networking (NDN) has emerged as a very viable paradigm for wireless networks, providing content availability features through stateful forwarding and innetwork caching. However, owing to its multicast communication model, it is susceptible to Interest Flooding Attacks (IFA) and broadcast storm problems, which lead to retransmissions, congestion, and degraded network performance. Existing mechanisms such as iFLAT, which constitute a heuristic-based filtering mechanism (using parameters such as Received Signal Strength (RSS), Interest satisfaction rate, blacklists, etc.), try to mitigate its effects. These mechanisms work for static environments but have reduced efficiency in adapting to changing morphisms of attacks with dynamic conditions within wireless networks. To overcome these dynamics, we propose RL-iFLAT, which stands for reinforcement learning-based adaptive forwarding strategy for IFA mitigation in NDN-based wireless networks. Unlike filtering mechanisms with fixed rules, RL-iFLAT learns and optimizes forwarding decisions dynamically through the application of Q-learning to modify Interest rate-limiting, forwarding probability, and anomaly detection thresholds. Our solution utilizes network traffic entropy, PIT occupancy rate, and Interest satisfaction rate as state parameters for making real-time forwarding decisions, effectively mitigating both inherent broadcast flooding and malevolent IFA. We evaluate RL-iFLAT on ndnSIM by implementing it in a Flying Ad hoc Network (FANET) utilizing Unmanned Aerial Vehicles (UAVs). From simulations, RL-iFLAT has been shown to reduce fake Interest forwarding by 40% while increasing the Interest satisfaction rate by 25% and PIT resource utilization by 30% when compared to iFLAT and other state-of-the-art approaches. This shows the efficacy of RL forwarding adaptation in improving.