METHODS FOR OPTIMIZING CONTAINER PLACEMENT IN DISTRIBUTED SYSTEMS

Abstract

Containerization is one of the key solutions in distributed service management due to its scalability and efficient use of resources. However, inefficient container placement in distributed computing environments can cause load imbalances between nodes, reduce overall system performance, and increase response times. This study aims to develop a method for dynamic container placement in distributed computing environments that does not require predictive models and improves the stability of web portals. To achieve this goal, existing approaches to containerization were analyzed, and an algorithm based on real-time resource monitoring, task prioritization, and availability constraints was developed. Research methods included test environment modeling, service deployment in Docker containers, Kubernetes orchestration, Prometheus monitoring, and performance metric analysis. The paper proposes an improved method for optimizing container placement based on real-time analysis of the current state of resources, considering CPU usage, memory, network traffic, and the importance of the tasks being performed. Unlike traditional approaches that use fixed rules or predictive models, the developed method uses adaptive responses to changing environmental conditions without prior estimates. The methodology ensures balanced load distribution and increases system stability. The results demonstrate that the proposed methodology reduced the average response time by 34 %, reduced the number of failures by 75 %, and increased resource utilization efficiency by 37 % compared to baseline algorithms (Round-Robin, static scheduling, predictive models). Testing results in a simulated load environment showed a 20–35 % reduction in task execution time and increased the efficiency of computing resource utilization. The practical significance of the work lies in the possibility of integrating the methodology into DevOps CI/CD cycles, where flexibility, scalability, and reliability are important. The approach is suitable for dynamic cloud environments, microservice architecture systems, automated deployment processes (CI/CD), and high-load web platforms. The methodology can serve as a basis for further research in intelligent service placement and self-learning resource management systems.