JESO: Reducing Data Center Energy Consumption Based on Model Predictive Control

Abstract

With the development of the Internet, the demand for data centers is growing dramatically. The cost of running a data center mainly comes from the huge electricity bill. Actually, IT (Information Technology) equipment and the HVAC (Heating, Ventilation, and Air Conditioning) system of the data center consume the majority of electricity. The existing energy–saving researches usually consider IT equipment or the HVAC system separately. But the energy consumption of HVAC is partially correlated with the running status of IT equipment. Taking methods to optimize the energy consumption of them jointly will generate more benefits. Therefore, we proposed JESO (Joint Energy Saving Optimization), a MPC (Model Predictive Control)-based method, to realize the joint energy-saving optimization of IT equipment and the HVAC system. We conducted extensive experiments based on generated transmission data and the HVAC system data from two real data centers. The experimental results demonstrated substantial energy reductions, achieving up to 51.67% in Fat-Tree and 45.03% in BCube network topologies. JESO outperforms separate optimizations of IT and HVAC systems, providing an additional energy reduction of 5.03% and 4.03% in these topologies, respectively.

Keywords

• Data center
• energy
• IT equipment
• HVAC