Research on Kubernetes resource optimization has expanded rapidly as containerized systems grow in complexity. This study proposes a deep learning–based approach for optimizing container resource configurations in managed Kubernetes clusters. The me...
Research on Kubernetes resource optimization has expanded rapidly as containerized systems grow in complexity. This study proposes a deep learning–based approach for optimizing container resource configurations in managed Kubernetes clusters. The method extracts historical CPU and memory usage metrics from containers within Pods and trains a combined CNN–LSTM model to capture complex temporal patterns in resource consumption. Using these patterns, the model predicts optimal CPU and memory requests and limits for the following seven days, enabling dynamic and workload-aware container configuration. We evaluate the approach on 13 production components representing diverse workload types, including deployment management services, stateful databases, and monitoring infrastructure. Experimental results show that the proposed model consistently outperforms heuristic, autoscaling, and metaheuristic optimization methods on the same clusters. Furthermore, it achieves statistically significant improvements in resource utilization. These findings demonstrate that the proposed deep learning approach is an effective solution for resource optimization in managed Kubernetes environments.