Experimental and Numerical Investigations of the Thermal Environment in Air-cooled Data Centers

Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Mechanical and Aerospace Engineering


Ezzat Khalifa

Second Advisor

Thong Dang


Reynolds Averaged Navier-Stokes, Temperature distribution

Subject Categories



The goal of the present work is to evaluate Reynolds Averaged Navier-Stokes (RANS) CFD as a tool to accurately predict the temperature distribution in air-cooled data centers with the aid of high quality test data. A data center research laboratory (RL) was constructed and heavily instrumented to measure power, flow rates, velocities, turbulent intensity, and temperature distribution around three simulated computer racks in the controlled RL environment. CFD models were developed and the results were validated against the experimental data measured in the RL. All CFD simulations were performed using a commercial CFD software package (ANSYS FLUENT 13) for the flow solver, and the computational grids were generated using the commercial grid-generation software GAMBIT 2.4.6.

Subsequently, a CFD parametric study was performed in order to illustrate the effect of various modeling factors on the accuracy of data center CFD simulations. The starting point of this parametric study was a comparison of the experimental data with the results of a baseline CFD case representing the current data center CFD simulation practice. Sequentially, several modeling factors were varied to study their effects on the CFD simulation results, e.g., turbulent boundary conditions, floor thermal boundary condition, grid resolution, geometry details, buoyancy, and tile flow model. The study showed that the strongest factors influencing and improving the CFD simulation results are the inclusion of correct tile flow model, buoyancy, and "realistic" turbulent boundary conditions. The remaining factors were found to be of secondary importance. A practical tile flow model "momentum source model" was developed to correct for the global values of both mass and momentum of the jets issuing from the perforated tiles and chassis exhaust doors. The inclusion of the momentum source compensates for the momentum lost when the flow from perforated openings is "smeared" over the entire opening to simplify the CFD representation. With the inclusion of momentum sources, buoyancy, and realistic turbulent boundary conditions, a reasonably accurate (~1.5 °C RMS error) temperature distribution can be obtained with grid size on the order of 4" typically employed in practical data center CFD simulations.


Open Access

This document is currently not available here.