CARBON-AWARE KUBERNETES SCHEDULING FOR ENERGY-EFFICIENT CLOUD-NATIVE COMPUTING ENVIRONMENTS

Abstract

The rapid expansion of cloud-native computing, containerized applications, microservices, artificial intelligence workloads, distributed analytics, Internet of Things platforms, and large-scale digital services has significantly increased the energy consumption and carbon impact of modern computing infrastructures. Kubernetes has emerged as a dominant orchestration platform for managing containerized workloads across heterogeneous clusters; however, conventional scheduling mechanisms primarily emphasize resource availability, performance, affinity constraints, and workload placement without explicitly considering temporal and spatial variations in electricity carbon intensity. As a result, computationally flexible workloads may execute in regions or time periods associated with comparatively carbon-intensive electricity even when lower-carbon alternatives are operationally feasible. This paper proposes a carbon-aware Kubernetes scheduling framework for energyefficient cloud-native computing environments. The proposed methodology integrates workload classification, Kubernetes telemetry collection, node-level energy profiling, regional carbonintensity information, renewable energy availability, service-level objectives, deadline sensitivity, resource requirements, application criticality, data locality, migration feasibility, and policy-driven scheduling. Workloads are categorized as latency-critical, deadlineconstrained, batch-oriented, deferrable, stateful, or migration-sensitive so that carbon optimization does not compromise application reliability. The framework employs a hierarchical decision process in which infeasible nodes are first removed according to resource, security, affinity, and operational constraints, after which eligible placement alternatives are evaluated according to carbon conditions, energy efficiency, utilization balance, latency, and workload urgency. Carbon-aware scheduling decisions are integrated with GitOpsbased policy governance, observability, controlled configuration management, and adaptive rescheduling. Representative experimental analysis across a simulated multicluster cloud-native environment demonstrates reductions in estimated carbon emissions, energy consumption, unnecessary workload movement, and low-utilization resource waste compared with conventional resource-centric scheduling. The results further indicate that workload flexibility is a major determinant of achievable carbon reduction, with batch and deferrable workloads providing greater optimization opportunities than strict latencycritical services. The proposed framework also supports production machine learning pipelines, enterprise APIs, Industrial Internet of Things services, Digital Twin applications, and modernized enterprise systems. The study demonstrates that carbon-aware Kubernetes scheduling can provide a practical foundation for sustainable, energy-efficient, policygoverned, and operationally reliable cloudnative computing.