Dynamic Power Management Through Multi-agent Deep Reinforcement Learning for Heterogeneous Systems DOI Open Access
Yiming Wang, Weizhe Zhang, Meng Hao

et al.

ACM Transactions on Architecture and Code Optimization, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 10, 2025

Power management and optimization play a significant role in modern computer systems, from battery-powered devices to servers running data centres. Existing approaches for power capping fail meet the requirements presented by dynamic workloads, situation becomes even more severe, given divergent energy efficiency of workloads on heterogeneous hardware platforms. Adaptively optimizing consumption presents great challenge systems. To tackle this challenge, we present machine learning based method improve system-level efficiency. We employ multi-agent deep reinforcement (MADRL) automatically explore relationship between long-term performance budget different types classic CPU-GPU Our framework equips each device with an agent, enabling decentralized control over its while maintaining centralized coordination maximize time applications within cap. evaluate our approach against state-of-the-art methods Experimental results show that improves average 8.5%. Additionally, is significantly stable compared heuristic approach.

Language: Английский

Dynamic Power Management Through Multi-agent Deep Reinforcement Learning for Heterogeneous Systems DOI Open Access
Yiming Wang, Weizhe Zhang, Meng Hao

et al.

ACM Transactions on Architecture and Code Optimization, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 10, 2025

Power management and optimization play a significant role in modern computer systems, from battery-powered devices to servers running data centres. Existing approaches for power capping fail meet the requirements presented by dynamic workloads, situation becomes even more severe, given divergent energy efficiency of workloads on heterogeneous hardware platforms. Adaptively optimizing consumption presents great challenge systems. To tackle this challenge, we present machine learning based method improve system-level efficiency. We employ multi-agent deep reinforcement (MADRL) automatically explore relationship between long-term performance budget different types classic CPU-GPU Our framework equips each device with an agent, enabling decentralized control over its while maintaining centralized coordination maximize time applications within cap. evaluate our approach against state-of-the-art methods Experimental results show that improves average 8.5%. Additionally, is significantly stable compared heuristic approach.

Language: Английский

Citations

0