Performance of a rack mountable cooling unit in an IT server enclosure. (1st June 2020)

• Record Type: Journal Article
• Authors: Moazamigoodarzi, Hosein; Pal, Souvik; Down, Douglas; Esmalifalak, Mohammad; Puri, Ishwar K.
• Is Part Of: Thermal science and engineering progress. Volume 17(2020)
• Publication Date: 2020-06-01
• Subjects: Rack mountable cooling units -- IT enclosure -- Flowrate mismatch -- ITE configurations -- Warm air leakage -- Temperature distribution; Heat engineering -- Periodicals; Heat engineering; Thermodynamics; Periodicals; 621.402
• DOI: 10.1016/j.tsep.2019.100395
• Languages: English
• ISSNs: 2451-9049

Abstract

Highlights: Enclosed rack with RMCU and separated chambers minimizes recirculation and bypass. Passive servers behave as porous ducts, allowing backward leakage of warm air. Cold chamber depth and IT load density and distribution affect RMCU performance. Mismatch between servers and RMCU flowrates is the main factor of RMCU performance. Abstract: Almost thirty percent of the power consumption in a data center (DC) is attributable to the cooling of IT equipment (ITE). This offers opportunities to reduce a DC's energy budget by considering alternatives to traditional cooling methods, such as perimeter or in-row air-conditioners, which suffer from inherent airflow deficiencies, e.g., hot air recirculation and cold air bypass. Due to these deficiencies, not enough cold air is supplied to the ITE, leading to hot spots that reduce equipment life significantly. Rack mountable cooling units (RMCUs), i.e., air-handlers mounted within the same enclosure containing the ITE, do not face this problem due to their proximity to servers. However, a proper understanding of airflow and temperature distribution inside an enclosure containing an RMCU has not yet been developed, preventing its widespread adoption. We investigate the temperature and airflow for various ITE configurations to determine the efficacy of an RMCU inside an enclosed rack. Experiments reveal that the principal factor influencing the temperature distribution is the leakage of warm air through passive servers, a hitherto unreported facet, which is dependent on the locations of these servers, IT load density and distribution, and cold chamber depth. This is also illustrated through a neural network model. The results help us to devise a design strategy to optimize ITE configurations with desirable temperature distributions, and to inform server workload assignments.