Studies on thermal management in green data centres

• Main Author: Chew, Xin Ni
• Other Authors: School of Mechanical and Aerospace Engineering
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/60356
• Institution: Nanyang Technological University

In today’s world, almost all businesses and trades have online platforms available. Following this trend, the number of data centres have been expanding tremendously. Companies providing essential and important services such as banking would require reliable and consistent data source. Hence, oversupplying of power is often the adopted approach to prevent failures; which in turn leads to depletion, wastage of natural resources and the high cost incurred. Therefore, it is of paramount importance to carefully examine and manage all data centres so as to avoid detrimental and undesirable scenarios. Existing studies were mostly done on room level cooling based on past technology, which is insufficient for the new blade servers; more efficient cooling at the server level can be explored. This project focuses on the generic layout of servers within a rack, airflow and thermal profiles within a server and how these factors would affect the overall airflow pattern and thermal profile within the rack. Dell™ PowerEdge™ R620 blade server is being used specifically in this case study. Computational Fluid Dynamic (CFD) models were built using the ICEPAK software. Detailed model of server was modeled to carry out simulations of airflow, heat transfer and thermal variations are studied. The models are depicted as close as possible to the actual servers so as to capture realistic datas. With the aim of understanding the complex airflow and thermal patterns in existing racks, each unit of the server modeled would eventually be represented by an equivalent reduced model so as to cut down on computational time and at the same time simplifying the rack-room modeling. Airflows and heat dissipation patterns would be predictable upon understanding of server-rack interaction. Parametric studies on consolidation of workloads from various pre-determined factors can be carried out.