Enhancing Network Efficiency: Empowering Intelligent Caching with Graph Neural Networks in Information-Centric Networking

Abstract

In the current era of ever-increasing data volumes and network traffic, efficient caching mechanisms play an important role in mitigating latency, managing network workload, and enhancing content delivery efficiency. This thesis explores the application of Graph Neural Networks (GNNs) in intelligent caching within Information-Centric Networking (ICN)-based environments, aiming to optimize content caching, maximize cache hit ratios, and improve overall system performance. The thesis first introduces a GNN-based caching strategy for ICN networks, leveraging GNNs to predict content popularity and make informed caching replacement decisions. Subsequently, it presents a GNN-based proactive caching placement strategy for ICN networks, capturing user preferences to optimize caching placement decisions in order to enhance the overall user experience. Furthermore, the thesis delves into intelligent caching with GNN-based Deep Reinforcement Learning (DRL) in Software-Defined Networking-based ICN (SDN-ICN) networks, utilizing a centralized GNN-Double Deep Q-Network (GNN-DDQN) agent to make proactive caching placement decisions for all network nodes. Lastly, it presents a fully distributed caching strategy, where each edge node maintains a Spatial-Temporal Graph Attention Network-Soft Actor-Critic (STGAN-SAC) agent to make proactive caching placement decisions in a three-tier edge network. The thesis aims to develop a comprehensive framework for intelligent caching utilizing GNNs, evaluating the effectiveness of the proposed strategies using synthetic and real-world datasets and simulations across various network topologies. The experimental results demonstrate advancements over state-of-the-art caching approaches.