Efficient Implementation of Potential-Flow Airflow Prediction for Data Centers

We present improvements to airflow prediction techniques for data centers, specifically within potential flow models. As a potential-flow model presents a simplified solution to the room airflow physics, additional approximations can be implemented to improve runtime without changing the accuracy of potential-flow. The improvements concentrate on two main components of current prediction methods: namely, the pre-processing task of automatic and efficient grid generation and post-processing task of capture index (CI) calculation. We propose a variable grid oriented around the objects in the room, creating cells with variable sizes (in width, height, and depth). We also show how CI calculations can be made more efficient through an understanding of the local nature of CI. An empirical study of sample data center layouts shows that these improvements can yield significant improvements in speed while maintaining a good level of accuracy.