A Lightweight Green-Power-Aware Online Scheduling Heuristic for Dynamic IoT-Cloud Systems

Abstract

Dynamic DAG-based applications are increasingly executed on heterogeneous IoT-cloud systems. These systems differ in speed, power characteristics, and energy-supply composition. Many schedulers still focus on makespan or total energy only. This is not enough in green computing environments because two schedules with similar energy use can draw very different amounts of non-renewable energy. This paper presents a lightweight green-power-aware online scheduling heuristic for dynamic IoT-cloud systems. Applications are modeled as directed acyclic graphs with delayed task visibility. For each ready task, candidate machines are scored by a locally normalized rule that combines expected finish time, DVFS-based total energy, short-term resource unavailability, and brown-energy exposure. We compare the proposed method with HEFT and an energy-oriented GREENHEFT baseline on dynamic Montage-05d and Epigenomics-100k workflows. Results are reported as the mean and sample standard deviation over five matched release seeds. Relative to HEFT, the proposed method reduces mean brown energy by 2.22% on Montage-05d and 47.79% on Epigenomics-100k, while increasing makespan by 15.89% and 3.15%, respectively. A compact three-point sensitivity analysis further shows the expected trade-off between the brown-energy weight and completion time. The results support a bounded conclusion: explicit brown-energy awareness can improve renewable alignment in online DAG mapping, but the environmental gains may require additional execution time or total energy.