SUMMARY

Thermal management of data centers remains a challenge because of their ever-increasing power densities and decreasing server footprints. Current lack of dynamic control over global provisioning, and local distribution of cooling resources, often results in wasteful cooling. These trends motivate the development of a reliable and energy efficient framework for allocation of cooling resources to meet thermal management requirements, while minimizing energy consumption and adverse environmental impact.
Computational fluid dynamics and heat transfer (CFD/HT) have been used extensively to model thermal transport and air flow in data centers. However, the significant computational costs and time associated with accurate room-level CFD/HT simulations for data centers renders such simulations impractical for real-time prediction and control. Nevertheless, developing an effective control framework for data centers to minimize their power consumption requires such real-time prediction of serever inlet and central processing unit (CPU) temperatures. It is proposed to develop neural network-based models trained on datasets generated from offline CFD/HT simulations for real-time prediction of temperature distributions and energy consumption. These predictions can then be used to optimize cooling power consumption, while keeping all the server CPUs below their maximum temperature limits. To this end, a physics-based and experimentally validated room-level CFD/HT model is developed. Numerical simulations are performed for several data center operating situations to generate the training datasets. The validated neural network model can then be used to minimize the total energy consumed by the cooling infrastructure, with the constraint that all servers operate within their prescribed temperature thresholds. The overall framework targets reducing energy consumption, without compromising server reliability.