Joint optimization of cooling parameters and workload distributions based on model predictive control for rack-based data centers

To satisfy the thermal environment requirements, approximately 40 % of energy consumption is used for cooling in data centers. The efficient cooling system and workload management can improve heat dissipation, thereby reducing energy consumption. This paper introduced a novel joint optimization strategy (JOS) for cooling parameters and workload distributions designed for rack-based cooling data centers. The primary objective is to improve overall energy efficiency while ensuring the safe operation of the data center. The proposed strategy considers the thermal interaction between information technology (IT) equipment and cooling devices, as well as the heterogeneity among different servers, achieving granular optimization of cooling parameters and workload distributions. The impacts of different cooling parameters optimization on the energy management and thermal management performance were analyzed. The energy-saving potential and temperature field control effect considering server heterogeneity for the proposed JOS were investigated. The results show that, compared to univariate optimization, simultaneous optimization of supplied cold air temperature and airflow rate can achieve at least 4.7 % in energy-saving. Furthermore, compared with independent control of the cooling system, the proposed JOS effectively addresses server overcooling, achieving a more uniform temperature distribution and further reducing the cooling energy consumption by 5 %. It is noteworthy that the strategy demonstrates high computational efficiency while ensuring energy efficiency.