CFD Simulation Studies of High Performance Computing (HPC) Facilities

Abstract

High performance computing (HPC) facilities consist of a large number of computer servers, which dissipate thermal energy in a data center. Efficient heat dissipation of these high density servers is a major concern to increase the life of electronic devices. Cooling of these devices involve thermal management strategies based on energy efficient cooling in a raised floor configuration, which depends on proper air flow distribution in an under-floor plenum, configuration of perforated tiles and arrangement of servers in racks. The current investigation was carried out with an objective to study air flow pattern inside plenum and to find out a better predictive tile model for air flow distribution through perforated tile. The full scale Computational Fluid Dynamics (CFD) model was used to predict flow over perforated tiles and temperature distribution over server racks inlet at different cold aisles. The governing equations (mass, momentum and energy) were solved to compute flow and temperature distribution in the computational domain. The RANS based k–e turbulence model was used for estimating turbulent kinetic energy (k) and turbulence dissipation rate (e). The under-floor blockages and CRAC locations were found to be significant parameters influencing flow distribution in different cold aisles. The CFD model was validated by carrying out experimental works. The deviation in flow rates obtained by CFD model was within 15 % with respect to experimental values for different rows in cold aisle. Cold aisle containment for over provisioned case of data centre was studied using modified body force model, which predicted better hot air entrainment and momentum rise of cold air as compared to porous jump model. The root mean square error (RMSE) of the temperature predicted by the modified body force model was 1.68 °C, whereas the RMSE predicted by porous jump model was 1.72 °C. Based on the CFD simulation studies, some energy saving opportunities were suggested for improving the thermal performance of the HPC facility.