A Survey on LLM-based Network Resource Optimization

Abstract

With the rapid proliferation of high-bandwidth and low-latency services such as virtual reality (VR), holographic communications, and large-scale Internet of Things (IoT), the complexity of network resource management has increased significantly. Network resource optimization plays a crucial role in improving throughput, reducing latency, and enhancing energy efficiency by enabling efficient utilization of limited wireless and wired resources. Conventional approaches, including rule-based methods, mathematical optimization, and reinforcement learning-based techniques, can achieve satisfactory performance in specific environments. However, they suffer from limitations such as poor generalization to dynamic environments, high modeling complexity, and difficulties in real-time decision-making. To overcome these limitations, recent studies have begun to explore network resource optimization based on Large Language Models (LLMs). This paper presents a comprehensive survey of LLM-based network resource optimization techniques. Existing studies are classified according to the role of LLMs, and the characteristics and limitations of each approach are analyzed.