METHOD OF ADAPTIVE SCALING OF CLOUD RESOURCES BASED ON REINFORCEMENT LEARNING
Abstract
The article considers the problem of adaptive scaling of cloud infrastructure resources under conditions of dynamic workload changes, uneven utilization of computational resources, and the necessity to ensure stable operation of information services. It has been determined that traditional approaches to horizontal and vertical scaling are mainly based on static threshold rules or predictive models that do not provide sufficient flexibility in a changing environment and do not take into account the multicriteria nature of the resource management process. The feasibility of using reinforcement learning methods for the formation of an adaptive cloud infrastructure management strategy is substantiated. The paper analyzes modern methods of automated resource scaling, decision-making models in cloud systems, and approaches to the use of reinforcement learning methods in workload optimization tasks. A method for adaptive scaling of cloud resources is proposed, based on the use of an intelligent agent that evaluates the current state of the infrastructure, analyzes workload parameters, request processing delays, processor resource utilization, and memory usage, and forms an optimal scaling strategy. A state model, reward system, and mechanism for selecting control actions for automated resource allocation have been developed. Within the experimental study, the proposed method was compared with classical automated scaling algorithms. The obtained results confirmed a reduction in request processing delays, improved resource utilization efficiency, and enhanced stability of cloud system operation. The practical significance of the work lies in the possibility of applying the proposed method for automated resource management in large distributed information systems and modern services.