Auto-Scaling Techniques for Container Workloads in Kubernetes Clusters

Abstract

This paper presents a comparative study of automatic scaling mechanisms in Kubernetes clusters. The objective of this study is to conduct a comparative analysis of various techniques for automating the scaling of containerized applications in Kubernetes clusters. The methodological foundation of the research comprises a systematic review and analytical processing of current scientific publications in the field. The work examines the architectural principles, key configuration parameters, and built-in limitations of traditional tools, including the Horizontal Pod Autoscaler, Vertical Pod Autoscaler, and Cluster Autoscaler. Particular attention is devoted to advanced solutions designed to enhance the adaptability and predictability of scaling. These include event-driven scaling using KEDA, high-efficiency node management with Karpenter, and the implementation of predictive strategies based on machine learning models. The scientific novelty of the study lies in the description of a comparative classification model of autoscaling techniques, which enables the formulation of clear recommendations for selecting the optimal strategy based on the type of workload: microservice web applications, big data processing pipelines, or resource-intensive machine learning tasks. The analysis suggests that to achieve high performance and resilience, it is advisable to combine various approaches — including horizontal, vertical, and cluster scaling — supplemented by heuristic or predictive methods. The findings will be valuable to DevOps engineers, cloud system architects, and researchers focused on optimizing operational performance and resource management in modern distributed environments.