Energy-Efficient Cloud–Edge Scheduling with QoS Preservation: Simulation and Testbed-Based Evaluation

Abstract

Energy-efficient and QoS-preserving scheduling is a major requirement in cloud–edge computing environments, where heterogeneous resources, dynamic workloads, and distributed data sources create complex performance–energy trade-offs. Although existing orchestration platforms provide automated deployment, scheduling, scaling, and workload management, many of them remain compute-centric and reactive. They commonly prioritize CPU availability, memory capacity, and load balancing while providing limited consideration for energy consumption, network latency, data movement cost, and service-level agreement satisfaction. As a result, conventional scheduling strategies may lead to excessive energy usage, higher latency, poor resource utilization, and increased SLA violations in cloud–edge infrastructures. This paper presents an experimental evaluation of an energy-efficient cloud–edge scheduling approach with QoS preservation. The proposed scheduling model supports intelligent task placement, workload scheduling, auto-scaling, VM/container migration, and workload consolidation across heterogeneous cloud, fog, and edge resources. The model integrates compute suitability, network condition, data locality, workload prediction, and Energy–QoS optimization to generate adaptive orchestration decisions. Unlike baseline schedulers that optimize isolated objectives, the proposed approach jointly minimizes energy consumption, latency, makespan, data transfer cost, and QoS violation while improving resource utilization and throughput. The performance of the proposed scheduling model is evaluated using multiple metrics, including total energy consumption, energy saving, average latency, makespan, SLA satisfaction rate, SLA violation rate, resource utilization, scheduling overhead, throughput, and data transfer cost reduction. The proposed model is compared with three related scheduling approaches: default Kubernetes scheduling, energy-aware scheduling, and ML-based resource scheduling. Experimental results show that the proposed approach achieves 32.15% energy saving, reduces average latency to 41.60 ms, improves SLA satisfaction to 98.10%, and increases resource utilization to 90.30%. Ablation analysis further confirms that task preprocessing, context modeling, workload prediction, multi-criteria decision-making, Energy–QoS optimization, and feedback learning each contribute progressively to the final improvement. Statistical significance testing validates that the observed gains are consistent and meaningful across repeated experimental runs.