Method of Granular Analysis of Shared Resource Utilization in a Kubernetes Cluster

Abstract

In modern computing environments, the Kubernetes platform has become the standard for automated deployment, scaling, and management of containerized applications. With the growing popularity of Kubernetes worldwide, there is an increasing need for accurate and efficient analysis of resource consumption. This is critically important to ensure stable performance, optimal infrastructure utilization, and economically justified capacity planning. However, resource management in Kubernetes is complicated by the dynamic nature of workloads and the interdependence of applications operating within a shared environment. The discovery establishes initial conditions for the analysis, considering the main types of resources: processor, random-access memory, disk space, and network traffic. For each resource, weighting coefficients are defined that reflect its relative importance in the context of different task executions. Additionally, each resource is divided into four states: allocated, reserved, utilized, and free. Such a division allows for a more detailed picture of the actual state of infrastructure usage and supports decision-making based on both technical and economic efficiency. The primary model focuses on evaluating the resource consumption of a single application at a specific point in time. Within this model, the relationship between reserved and actually utilized resources is examined, enabling the identification of excessive reservation or, conversely, insufficient provisioning. The model forms the foundation for a basic understanding of the application's behavior in the cluster and allows for initial diagnostics of inefficiencies. Further model refinement incorporates changes in resource usage over time. The behavior of a single application is analyzed throughout a specified period, which opens the possibility to identify long-term trends, gradual resource leakage, peak loads, or uneven consumption. This approach significantly improves the accuracy of assessing application performance and supports informed decisions regarding scaling or setting constraints. The final level of the model provides for analyzing resource usage by multiple applications simultaneously over a given period. This allows consideration of the mutual influence between applications, competition for shared resources, and overall environment load. As a result, a comprehensive picture of resource balance is formed, which serves as the basis for intelligent cluster policy planning, workload placement optimization, and service stability assurance. The proposed approach to granular analysis of resource utilization in Kubernetes clusters is a promising research direction. It opens wide opportunities for further development of forecasting models, automated resource management, and the construction of adaptive monitoring systems capable of independently responding to changing cluster loads. The conclusions of the study emphasize the importance of integrating such methods to enhance the efficiency of modern cloud infrastructure operations.