Data Center Heat Distribution Modeling Using Onboard Sensors

Abstract

Uneven cooling in data centers can be detrimental to server and cooling unit performance. To maintain a desired cooling environment, an appropriate thermal management strategy is necessary such as creating effective visual thermal maps to characterize cold air and hot air distribution. Numerically analyzing heat distribution models along with prescribed environmental sensor measurements as boundary conditions generates thermal maps. Enabling reliable and real-time heat distribution monitoring suggests the need for computationally efficient and accurate models and optimized sensor placements.This study presents a customized or Custom heat distribution modeling which considers internal server heat, and applies onboard server sensor measurements. With the Navier-Stokes model using external and onboard sensor measurements as the baseline, the Laplacian-convection-diffusion (LCD) and the Custom models as well as the onboard versions of all models provide comparable mean absolute errors or MAEs. The Custom model produces competitive results especially in using both external and onboard sensors where cooling vents are located. The Custom model converges faster than Navier-Stokes but slower than LCD. In essence, the Custom model using onboard sensors reasonably provides the most computationally accurate, efficient and attainable solution suitable for real-time monitoring and effective in reducing electrical costs and in avoiding any downtime in the data center.