Impact of fan pressure change and rack layout on data center thermal performance

Abstract

Data center is a housing facility for computing related devices and associated components such as computer servers and telecommunication equipment. Due to the rapid progression of technology with high speed processors in internet application, data centers have been the focus of thermal management efforts. The increased server heat density in a server rack is major concern for data center operators. Effective and efficient removal of server heat is necessary to maintain safe working environment inside the data center. Therefore there is a need for data center thermal management. Although a number of experimental and numerical studies have focused on fluid flow and heat transfer within data centers, systematic studies of server rack layout, the effect of server fans and validation of numerical model with laboratory prototype are missing from the literature. In this thesis work, we have addressed and provided an efficient solution to this issue by changing rack layout in a data center. We have analysed the effect of server fan pressure rise, and compared the conventional data center layout with an alternative “S-Pod” layout using computational fluid dynamics simulations. We have presented the theoretical basis for scaling down a data center to a lab-scale prototype, and then validated numerical models with experimental results on this prototype. We have also studied effectiveness of passive rear door heat exchanger modelled using porous media approach. Conventional data center showed better thermal performance for the server rack fan pressure of 10 Pa while S-Pod layout showed better thermal performance at 5 Pa. Similar thermal profiles are observed for conventional layout and S-Pod layout having an additional 20% heat load. The use of Richardson’s number in dimensional analysis has been found very promising in building an experimental prototype of data center. The improvement in performance indices (SHI, RHI, RCI and RTI) in S-Pod layout (scaled down data center) has been observed to be 10.5%, 4.5%, 6.2% and 3.8% based on simulation result and 5.9%, 3%, 3.3% and 3.5% based on experimental result. Through experiments and CFD simulations, we demonstrated that during various transients such as air-conditioning failure and a load fluctuations, S-Pod layout results in consistently (relatively) superior performance. Results indicate an optimum value of fan pressure change and an additional 20% heat load bearing capacity in S-Pod layout because of confined distribution of air flow. Considering all the indices cumulatively, S-Pod layout is almost 5-10% more efficient as compared to conventional layout. These studies quantify the improvement in performance, and are significant for research and data center designers.