A Deep Reinforcement Learning-Based Elastic Scaling Method for Kubernetes Container Cloud

To achieve on-demand allocation of physical resources in container clouds and improve the reliability of elastic scaling, this paper proposes a container elastic scaling method based on reinforcement learning algorithms. The method first establishes a Markov Decision Process model for container cloud scaling, considering real-time scaling requirements and a service priority-based resource allocation strategy. Subsequently, a container scaling algorithm is designed using a Deep Q-Learning Network to obtain the optimal resource allocation. Simulation results comparing the proposed method with existing elastic scaling approaches demonstrate that, regardless of whether the scenario involves periodic requests or burst requests, the proposed method shows significant superiority in terms of business continuity and service reliability. Furthermore, the analysis of resource occupancy for services with different priorities indicates that this method can prioritize the resource requests of highpriority services in resource-constrained situations, thereby facilitating intelligent tiered resource management for container clouds.