Air flow management inside data centres

A data centre can be defined as an infrastructure facility that houses file servers, processors and other computer equipment, along with a standby power supply. These servers are kept inside cabinets and those cabinets are called racks. These racks are located close to each other inside a data centre to form rows. These rows are located front to front and back to back to form the aisles. These aisles could be used to supply the chilled air and also to provide room for operational purposes. Data centres are now widespread due to the high demand of infrastructure requirements, such as the network to operate Internet services. In this thesis, research is focused on the air cooling method, a popular method of cooling that is used to cool many data centres. The aim of this thesis is to understand the capabilities and limitations of Computational Fluid Dynamics (CFD) analysis of cooling air flow in data centres. The data centre components, which are the server blade and rack, have been simulated in order to study the environmental conditions (temperature, pressure and velocity fields) inside the data centre; as such, CFD analysis has been carried out at server, rack and room levels. The proposed method of a porous media model has been implemented to simulate servers and racks and has been tested and validated through corresponding experiments. It is shown from the results that the porous media model provides good agreement with experimental data of an actual case at the server level. The server racks have been simulated as a porous media with different permeability values in each direction (x ,y, z). In addition, a 3-dimensional CFD model has been used to explore the performance of three different room level cooling strategies based on the aisle containment (cold and hot aisle containments) and back door cooler. It is shown that using either cold or hot aisle containment within a data centre provides significant improvement inside the data centre with respect to temperature distribution and the avoidance of hot spots. Finally, the power input to the computer room air conditioning (CRAC) unit has been analysed for different cooling configurations when assuming the Coefficient of Performance (COP) of either direct expansion CRAC unit or a chiller system. Furthermore, the comparison between active and passive back door coolers has been done to evaluate the power consumption in the CRAC unit. It is shown that the supply temperature inside the data centre has a significant effect on the CRAC power input (compressor work) of the DX CRAC unit. With respect to comparison between the active and passive back door coolers, it has been found that the reduction of the CRAC unit load is higher when using the active back door cooler compared to the passive back door cooler, so the active back door cooler is better than passive back door cooler with respect to reduction of load on CRAC unit.