Physical Models in Data Center Airflow Simulations
The trend of increasing functionality of electronics with a reduction in size has caused a rapid increase in the volumetric heat generation of today’s equipment. This problem is compounded by the vertical stacking of such components in tall enclosures, called racks. The organization of these racks into large infrastructure facilities, or data centers, generates enough heat to require a room-level cooling strategy. The total power dissipated in current data centers can be as large as several MW. Since all the heat generated must be removed, a systematic thermal management methodology is required to ensure efficient, reliable and safe operating conditions. The mathematical description of the airflow and heat transfer characteristics involves a range of length scales, all of which are infeasible to approach simultaneously. In this study, a modeling framework for data centers is investigated, with emphasis on the physical models employed in numerical simulations. Computational fluid dynamics (CFD) models are presented to develop a unit cell, or a minimum sized model which is representative of these facilities. A unit cell architecture is a useful design tool for the evaluation of tomorrow’s cooling strategies. A premium on floor space may result in oddly shaped facilities, so a need exists for a common basis of comparison. The flow patterns inside a data center typically fall into the regime of turbulent mixed convection. The choice of turbulence model employed in a Reynolds-averaged Navier-Stokes (RANS) type simulation is examined using a commercial code. Comparisons are made with indoor airflow simulations of office spaces and auditoria because of the similarity of office spaces and auditoria because of the similarity in velocities and length scales. Results shows up to 20% variation in temperature predictions can occur between various commercially-implemented turbulence models.