Inviscid and Viscous Numerical Models Compared to Experimental Data in a Small Data Center Test Cell

Abstract

Both localized power densities and overall power consumption within the data center continue to rise, following the same upward trend as the information technology (IT) equipment stored within the data center. Air cooling this increasing power has proved a significant challenge at both the IT equipment and data center level. In order to combat this challenge, computational fluid dynamics and heat transfer (CFD/HT) models have been employed as the dominant technique for the design and optimization of both new and existing data centers. This study is a continuation of earlier comparisons of CFD/HT models to experimentally measured temperature and flow fields in a small data center test cell. It compares an inviscid model, a laminar flow model, and three turbulence models to six sets of experimentally collected data. The six sets of data are from two different IT equipment rack power dissipations using three different layouts of perforated tiles. Insight into the location of the deviation between the different CFD/HT models and experimental data are discussed, along with the computational effort involved in running the models. A new grid analysis was performed on the different CFD/HT models in order to try to minimize computational effort. The inviscid model was able to run with a smaller grid size than the viscous models and even for the same size grid was found to run 30% faster than the fastest viscous model. Due to both the reduced grid size and computational effort (due to the simpler equation set), the inviscid model ran over thirty times faster than the next fastest model. The fact that the inviscid model ran the fastest is not surprising, however what was not expected is that the inviscid model was also found to have the smallest deviations from the experimental data for all six of the cases. This is most likely due to the arrangement of the data center test cell with the relatively few high velocity air jets and large open space around the IT equipment. More tightly packed data centers with higher air velocities and turbulent mixing conditions will certainly produce different results than those found in this study.