Scalable Energy-Aware Task Scheduling on Heterogeneous Voltage Frequency Islands: From Optimal ILP to Practical Heuristics With Real-World Validation
IEEE Access · 2026 · 14 · 84196-84213
DOI: 10.1109/access.2026.3695239 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Sanchit und 5 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This paper presents a comprehensive approach to energy-efficient DAG scheduling on Voltage Frequency Islands (VFI) architectures, combining optimal Integer Linear Programming (ILP) formulation with a novel Energy-Aware List Scheduling with Dynamic Slot Reuse (EALS-DSR) heuristic. The work addresses the critical challenge of minimizing makespan and overall energy consumption through intelligent task allocation, voltage/frequency assignments, and scheduling optimization. Our contributions include: 1) a detailed ILP formulation with theoretical guarantees for optimal solutions, 2) the EALS-DSR heuristic providing near-optimal solutions with polynomial complexity O(n2p), 3) comprehensive experimental validation on LUD, FFT, and Random DAGs, and 4) real-world molecular dynamics simulation case study. To minimize makespan and overall energy consumption, a Directed Acyclic Graph (DAG) can be scheduled on voltage frequency islands by first splitting the available processing units into numerous islands with different voltage and frequency levels. This paper presents a comprehensive approach combining Integer Linear Programming (ILP) with a novel Energy-Aware List Scheduling with Dynamic Slot Reuse (EALS-DSR) heuristic. The proposed methodology includes four main components: (a) DAG modeling, (b) job allocation based on slots, (c) voltage/frequency assignments to each slot, and (d) scheduling optimization. We extend the original ILP formulation with theoretical analysis, comprehensive experimental validation, and real-world case studies including molecular dynamics simulation. Results demonstrate that the ILP approach improves performance by 10-24% over existing methods, while EALS-DSR achieves 15-25% energy savings with excellent scalability for large-scale problems. The molecular dynamics case study shows 37.4% energy reduction and $3.86\times $ efficiency improvement, validating practical applicability.
Zitieren
APA
Sanchit, Sakshi, Singal, M., Dhaked, D. K., V., L. N., & Singh, J. (2026). Scalable Energy-Aware Task Scheduling on Heterogeneous Voltage Frequency Islands: From Optimal ILP to Practical Heuristics With Real-World Validation. IEEE Access, 14, 84196-84213. https://doi.org/10.1109/access.2026.3695239
Vancouver
Sanchit, Sakshi, Singal M, Dhaked DK, V. LN, Singh J. Scalable Energy-Aware Task Scheduling on Heterogeneous Voltage Frequency Islands: From Optimal ILP to Practical Heuristics With Real-World Validation. IEEE Access. 2026;14:84196-84213. doi:10.1109/access.2026.3695239
BibTeX
@article{Sanchit2026,
author = {Sanchit and Sakshi and Singal, Maanas and Dhaked, Dheeraj Kumar and V., Lakshmi Narayanan and Singh, Jagpreet},
title = {Scalable Energy-Aware Task Scheduling on Heterogeneous Voltage Frequency Islands: From Optimal ILP to Practical Heuristics With Real-World Validation},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {84196--84213},
doi = {10.1109/access.2026.3695239}
}
EUVIMED