TY - GEN
T1 - Jointly Optimizing the IT and Cooling Systems for Data Center Energy Efficiency based on Multi-Agent Deep Reinforcement Learning
AU - Chi, Ce
AU - Ji, Kaixuan
AU - Marahatta, Avinab
AU - Song, Penglei
AU - Zhang, Fa
AU - Liu, Zhiyong
N1 - Publisher Copyright:
© 2020 ACM.
PY - 2020/6/12
Y1 - 2020/6/12
N2 - With the development and application of cloud computing, the increasing amount of data centers has resulted in huge energy consumption and severe environmental problems. Improving the energy efficiency of data centers has become a necessity. In this paper, in order to improve the energy efficiency of both IT and cooling systems for data centers, a model-free deep reinforcement learning (DRL) based joint optimization approach MACEEC is proposed. To improve the cooperation between IT and cooling system while handling the high-dimensional state space and the large hybrid discrete-continuous action space, a hybrid AC-DDPG multi-agent structure is developed. A scheduling baseline comparison method is proposed to enhance the stability of the architecture. And an asynchronous control optimization algorithm is developed to solve the different responding time issue between IT and cooling system. Experiments based on real-world traces data validate that MACEEC can effectively improve the overall energy efficiency for data centers while ensuring the temperature constraint and service quality compared with existing joint optimization approaches.
AB - With the development and application of cloud computing, the increasing amount of data centers has resulted in huge energy consumption and severe environmental problems. Improving the energy efficiency of data centers has become a necessity. In this paper, in order to improve the energy efficiency of both IT and cooling systems for data centers, a model-free deep reinforcement learning (DRL) based joint optimization approach MACEEC is proposed. To improve the cooperation between IT and cooling system while handling the high-dimensional state space and the large hybrid discrete-continuous action space, a hybrid AC-DDPG multi-agent structure is developed. A scheduling baseline comparison method is proposed to enhance the stability of the architecture. And an asynchronous control optimization algorithm is developed to solve the different responding time issue between IT and cooling system. Experiments based on real-world traces data validate that MACEEC can effectively improve the overall energy efficiency for data centers while ensuring the temperature constraint and service quality compared with existing joint optimization approaches.
KW - cooling system
KW - data center
KW - deep reinforcement learning
KW - energy efficiency
KW - multiagent
KW - scheduling algorithm
UR - http://www.scopus.com/inward/record.url?scp=85088516660&partnerID=8YFLogxK
U2 - 10.1145/3396851.3402658
DO - 10.1145/3396851.3402658
M3 - Conference contribution
AN - SCOPUS:85088516660
T3 - e-Energy 2020 - Proceedings of the 11th ACM International Conference on Future Energy Systems
SP - 489
EP - 495
BT - e-Energy 2020 - Proceedings of the 11th ACM International Conference on Future Energy Systems
PB - Association for Computing Machinery, Inc
T2 - 11th ACM International Conference on Future Energy Systems, e-Energy 2020
Y2 - 22 June 2020 through 26 June 2020
ER -
