The Role of LLMs in Predictive and DecisionSupport Mechanisms for Kubernetes Auto-Scaling
AbstractLarge Language Models can enhance Kubernetes- based auto-scaling processes by acting as both predictive analysis tools and decision-support mechanisms. In Kubernetes, auto- scaling is typically managed using Horizontal Pod Autoscaler and Vertical Pod Autoscaler, while Cluster Autoscaler handles scaling at the node level. However, traditional metric-based scaling approaches often fall short when dealing with dynamic and unpredictable workloads. This mini-review explores the integration of LLMs into Kubernetes auto-scaling. In predictive scaling, LLMs can analyze historical workload data to forecast demand and optimize scaling decisions. As decision-support tools, they assist in diagnosing system failures, detecting anomalies, and improving key operational factors such as energy efficiency and resource allocation. The review categorizes LLM-enhanced auto- scaling solutions into traditional metric-based approaches, machine learning-driven prediction models, and hybrid systems, addressing key challenges such as latency, scaling sensitivity, and computational costs. By incorporating LLMs into Kubernetes auto-scaling workflows, systems can become more autonomous and intelligent, leading to improved efficiency and resilience in cloud-native environments.