An Optimized GNN-Based Caching Scheme for SDN-Based Information-Centric Networks.

Information-Centric Networking (ICN) has recently attracted much attention due to in-network caching and named-based routing. Caching has played a critical role in the era of proliferating network traffic. More importantly, effective caching can improve content delivery and optimize network efficiency. This paper aims to improve cache performance by maximizing cache hit ratio, minimizing content delivery latency, and path stretching in the Software-Defined Networking-ICN (SDN-ICN) context. We first propose a statistical model to generate users' preferences for different contents, which contains many essential ingredients observed in real-world scenarios. Next, we built a Graph Neural Network-Deep Reinforcement Learning (GNN-D RL) agent to generate caching decisions for each node (and at every time step) based on users' content request history. Simulation results show that the proposed caching scheme can reach a 12.3 % higher cache hit ratio, 6.5 % lower average latency time and 6.5 % lower average path stretch compared to the state-of-the-art caching strategy.