ABSTRACT The dynamic nature of cloud‐native applications necessitates robust resource elasticity to meet performance objectives while optimizing costs. However, conventional container orchestration systems often rely on reactive, static‐threshold mechanisms and centralized control architectures, which lead to inefficient resource utilization and scalability bottlenecks. This paper presents EffiScale, a novel, self‐adaptive framework for orchestrating container elasticity in cloud environments. EffiScale introduces a decentralized microservice architecture, extending the MAPE‐K model, that integrates four core innovations: (1) a decentralized control plane governed by a Byzantine fault‐tolerant consensus protocol, eliminating single points of failure and ensuring resilient coordination; (2) a hybrid scaling orchestration engine that frames the choice between vertical and horizontal scaling as a multi‐objective optimization problem, enabling granular and cost‐effective resource allocation; (3) a predictive, risk‐aware thresholding mechanism that leverages an ensemble of machine learning models to forecast workloads and dynamically adjusts scaling triggers based on both the prediction and its associated uncertainty; and (4) a federated knowledge base for continuous, collaborative refinement of scaling policies across distributed controllers. Experimental evaluation on a multi‐node cluster with a stateless Nginx web application demonstrates that EffiScale can reduce response times, increase throughput, and lower resource usage compared to state‐of‐the‐art solutions, highlighting its effectiveness in managing dynamic containerized workloads.
Ahmadpanah et al. (Wed,) studied this question.