Potential Flow Modeling for Fast Data Center Thermal Simulation

Data-center-airflow-and-energy analyses by computational fluid dynamics (CFD) can be conducted at various levels of speed and accuracy as required by the application and simulation goals. For example, rapid prototyping in the early design, which normally involves optimization for high-level design goals, favors computational speed over absolute accuracy. Conversely, calibrating a model against experimental measurements in an existing data center requires greater accuracy. Full CFD has been shown to be sufficiently accurate for general data-center applications when appropriate modeling practices were followed. In this paper we compare the accuracy and speed of potential flow modeling (PFM) to full CFD as applied to data-center examples featuring three common cooling architectures. We observe that PFM can provide reasonable predictions for many traditional-downflow-cooling and row-cooling architectures either with or without traditional (aisle-depth) containment. PFM does less well with row-based-cooling applications and those featuring strong jet flow like that associated with front-discharge-cooling-unit supplies. However, when PFM can be employed, we find that it is approximately up to almost 40 times faster than our (already very fast) benchmark full CFD with identical computational grids and boundary conditions.