A Numerical and Experimental Study of Airflow in Data Centers

Abstract: Airflow management is crucial for the performance of cooling systems in data centers. The amount of energy consumed by data centers is huge and a large amount is related to the cooling. In attempts to develop energy efficient data centers, numerical methods are important for several reasons. Experimental procedures are more expensive and time consuming but when done carefully, experiments provide trustful results that can be used to validate simulations and give additional insights. Numerical methods in their turn have the advantage that they can be applied to proposed designs of data centers before they are built and not only to already existing data centers.In this study, Computational Fluid Dynamics (CFD) is used to study the airflow in data centers. The aim is to use an experimentally validated CFD model to investigate the effects of using different designs in data centers with respect to the performance of the cooling systems. Important parameters such as quality of the computational grid, boundary conditions and choice of turbulence model must be carefully considered in order for the results from simulations to be reliable.In Paper A, a hard floor configuration where the cold air is supplied directly into the data center is compared to a raised floor configuration where the cold air is supplied into an under-floor space instead and enters the data center through perforated tiles in the floor. In Paper B, the performance of different turbulence models are investigated and velocity measurements are used to validate the CFD model. In Paper C, the performance of different cooling systems is further investigated by using an experimentally validated CFD model. The effects of using partial aisle containment in the design of data centers are evaluated for both hard and raised floor configurations.Results show that the flow fields in data centers are very complex with large velocity gradients. The k − ε model fails to predict low velocity regions. Reynolds Stress Model (RSM) and Detached Eddy Simulation (DES) produce very similar results and based on solution times, it is recommended to use RSM to model the turbulent airflow in data centers. Based on a combination of performance metrics where both intake temperatures for the server racks and airflow patterns are considered, the airflow management is significantly improved in raised floor configurations. Using side covers to partially enclose the aisles performs better than using top covers or open aisles.