Modeling of High Power Density Data Centers

Jeff Rambo

Data centers are large infrastructure facilities that house data processing and storage equipment.  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 2 m tall enclosures, called ‘racks’.  Today’s racks may dissipate up to 10 kW and with data facilities reaching up to 50,000 ft2, the total energy dissipated is on the order of several MW.  Since all the heat generated must be removed from the electronic components, a facility-level cooling scheme is required.  The cost to power a data center alone can be on the order of millions of dollars a year, with the cost to provide adequate cooling not far behind.  The need for improved characterization of data centers is motivated by the inadequacies of simple energy balances to provide an accurate modeling framework by which future facilities can be design.  An efficient and reliable cooling strategy is therefore required by the end user since most data centers are required to operate continuously

The predominate cooling strategy in today’s data centers is to arrange the racks in alternating directions of exhaust air to form ‘hot’ and ‘cold’ aisles, see Fig. 1.  A computer room air conditioning (CRAC) unit supplies a raised floor plenum with cold air, which is drawn into the racks through perforated tiles.  The hot exhaust air is collected from the upper portion of the data center by the CRAC unit to complete the airflow loop.

Fig. 1  Airflow schematic of a raised floor plenum data center

Objectives & Results

The goal of this project is to provide a modeling frame and design methodology for tomorrow’s data centers through computation and experimental methods:

1. A multi-scale approach through computational fluid dynamics (CFD) is being used to predict the complex velocity field and sub-rack scale thermal responses.  Parametric studies have shown significant temperature variations in the vertical direction of the rack and rack flow rate deviations associated with various rack locations.  A computed temperature field can be seen in Fig. 2 and Fig. 3 shows the complex velocity inside a data center.

2. A unique 1000 ft² data center laboratory facility is being constructed in cooperation with industry partners to provide experimental justification to the computational results.  The data will also be used to assess the performance of various turbulence models in data center airflow simulations.

Fig. 2. Temperature field [°C] from a vertical cross section

Fig. 3. Velocity vector map [m/s] from a horizontal cross section