Energy-constrained workflow scheduling for DVFS-enabled heterogeneous systems in cloud computing environments

Workflow scheduling in heterogeneous multiprocessor systems becomes substantially harder when a global energy budget must be satisfied alongside performance goals. Existing energy-constrained schedulers share a common weakness: task-level energy allocation is computed independently of processor assignment, causing pre-assigned budgets to misalign with the energy-frequency characteristics of the processors eventually selected. Greedy runtime methods compound this by exhausting slack on early tasks, starving critical tasks later in the schedule. This paper presents Energy Restricted Workflow Scheduling (ERWS), a three-phase heuristic for dynamic-voltage-and-frequency-scaling-enabled heterogeneous systems that minimizes makespan under a hard global energy constraint. ERWS prioritizes tasks using upward ranking, distributes slack energy in proportion to each task’s minimum energy requirement, and selects processor-frequency pairs by earliest finish time subject to per-task energy limits. The method is evaluated on six real-world scientific workflows and compared against five state-of-the-art energy-constrained scheduling methods. Under the tightest budget tested, ERWS reduces normalized schedule length by an average of 39.3 % over the greedy baseline and 5–9 % over existing static pre-assignment methods; gains persist, though with narrower margins, as the budget relaxes. Friedman and Wilcoxon signed-rank tests confirm statistical significance across all workflows and all 30 pairwise comparisons.