Predictive Model Development and Validation for Raised Floor Plenum Data Center

TLDR: This paper develops a data-informed, experimentally validated and computationally inexpensive system level predictive tool that can forecast data center behavior for a broad range of operating conditions and expects that this model can form an important building block in a future intelligent, increasingly automated data center environment management systems.

Abstract: With the explosion in digital traffic, the number of data centers as well as demands on each data center, continue to increase. Concomitantly, the cost (and environmental impact) of energy expended in the thermal management of these data centers is of concern to operators in particular, and society in general. In the absence of physics-based control algorithms, computer room air conditioning (CRAC) units are typically operated through conservatively predetermined set points, resulting in suboptimal energy consumption. For a more optimal control algorithm, predictive capabilities are needed. In this paper, we develop a data-informed, experimentally validated and computationally inexpensive system level predictive tool that can forecast data center behavior for a broad range of operating conditions. We have tested this model on experiments as well as on (experimentally) validated transient computational fluid dynamics (CFD) simulations for two different data center design configurations. The validated model can accurately forecast temperatures and air flows in a data center (including the rack air temperatures) for 10–15 min into the future. Once integrated with control aspects, we expect that this model can form an important building block in a future intelligent, increasingly automated data center environment management systems.