Thermal zones for more efficient data center energy management

Abstract

The optimization of cooling resources within data centers (DCs), which includes the localized provisioning of cooling power with respect to the heat dissipation, can lead to significant reduction of energy consumption [1]. While such matching is rather straightforward for the entire DC, on a more granular level however, the large range of heat densities of today's IT equipment has made this task much more challenging. One complication comes from the fact there is little operational information, for example, which physical areas (or zones) are supplied by the different air conditioning units (ACUs) so that local cooling demands can be met [2]. Evidently, such zones are not only determined by the placement and air flow produced by each ACU but also by the locations of perforated tiles throughout the DC and many other parameters. Here we present a general methodology for determining such zones. Specifically, we use a simplified physics model to calculate the air flow velocity field within the DC [3,4,5]. The novelty of the approach is that the boundary conditions are obtained from real-time measurements. Once the air flow pattern is calculated, we deploy an efficient tracing algorithm, which follows the air from/to each area of the DC back to the originating/returning ACU. The trace connects a specific ACU with its related zone. Each zone can be associated with the actual efficiency of the corresponding ACU by considering the respective utilization of the units (heat removed / nominal heat load removal capacity) or coefficient of performance (COP=heat removed / power consumption for ACU fans). Both COP and ACU utilization can be obtained from real-time temperature and air flow measurements. The paper will also discuss a software tool, which helps operators to use this information for improving DC energy efficiency. ©2010 IEEE.