Experimental investigation of design and performance of data centers

Data centers will be the computational hub of the next generation. Hosting business and mission-critical applications demand a high degree of reliability and flexibility. Deployment of large number of high powered computer systems in very dense configurations in racks in a data centers will result in very high power densities at room level. Managing high power levels in a data center with cost effective reliable cooling solutions is essential for reliability and uptime. Energy consumption of data centers can also be severely increased by over-designed air handling systems and rack layouts that allow the hot and cold air streams to mix. In this paper, we present experimental results from a design of experiments conducted at a production data center facility. The "smart" data center includes a pervasive monitoring layer that enables controlled deployment of air conditioning resources based on demand. Temperature distribution across the data center is measured by a large distribution of sensors at different computer room air conditioning (CRAC) unit air flow and supply temperature configurations. Dimensionless parameters in the form of Supply Heat Indices (SHI) are calculated, based on rack inlet, outlet and CRAC unit supply temperatures, at rack level for validation purposes. Experimental data show that SHI provides a simple and powerful tool to understand the convective heat transfer and fluid flow in diverse regions inside the data center. Analysis based on supply heat index is carried out to understand the optimization of relative air flow distribution among heterogeneous rack heat loads. The index is also used to investigate CRAC unit influences and rack air flow. Results show that these parameters not only provide an invaluable tool to understand convective heat transfer in large data centers but also suggest means to improve energy efficiency in data centers.