Dynamic thermal management of data centers

The usage of computers and internet in this day in age is considerably increasing on a universal level, which has driven major advancements in the semiconductor and packaging industry to provide faster computing capabilities in smaller packages. This is causing drastic increases in heat dissipations via the different Information Technology (IT) equipment. It is for this reason why cooling data centers typically requires over a third of the total facility energy consumption, and finding ways to control it and optimize it is of great importance. The work presented in this dissertation focuses on the two core aspects of cooling data centers mentioned above; energy efficiency and control.;The energy efficiency portion of the work focuses on cold aisle containment. Although containment within itself reduces power consumption by separating the cold and hot air streams, further possibilities of reduction in power consumption can be attained through using provisioned leakage into the containment which allows for lowering fan power consumption.;The control portion of the work focuses on dynamic cooling. Dynamic cooling has been proposed as one approach for enhancing the energy efficiency of data center facilities. This dissertation takes the first steps towards dynamic (transient) modeling of data centers using Computational Fluid Dynamics (CFD). Effects of variations in power and airflow on the data center environment are shown in time, as well as the effects of modeling complexities such as IT equipment thermal mass. It is shown through the results the importance in accounting for IT equipment thermal mass as it dictates how slowly or quickly the facility temperature rises or falls.;Although thermal mass is proven to be extremely significant in the transient response of data centers, modeling it in CFD is highly computationally expensive. To simplify the problem, simple energy balance principles and extensive experimental measurements are used to extract the thermal mass of two 2 RU servers and the obtained values varied from 11 – 12 kJ/K. The obtained thermal mass is used to develop compact models in CFD that can accurately represent the rise in temperature difference across the server in time while significantly saving on computational time.