UC Berkeley

Information technology (IT) is becoming increasingly pervasive throughout society as more data is digitally processed, stored, and transferred. The infrastructure that supports IT activity is growing accordingly, and data center energy demands have increased by nearly a factor of four over the past decade. Data centers house IT equipment and require significantly more energy to operate per unit floor area than conventional buildings. The economic and environmental ramifications of continued data center growth motivate the need to explore energy-efficient methods to operate these buildings. A substantial portion of data center energy use is dedicated to removing the heat that is generated by the IT equipment. Using economizers to introduce large airflow rates of outside air during favorable weather could substantially reduce the energy consumption of data center cooling. Cooling buildings with economizers is an established energy saving measure, but in data centers this strategy is not widely used, partly owing to concerns that the large airflow rates would lead to increased indoor levels of airborne particles, which could damage IT equipment. The environmental conditions typical of data centers and the associated potential for equipment failure, however, are not well characterized. This barrier to economizer implementation illustrates the general relationship between energy use and indoor air quality in building design and operation. This dissertation investigates how building design and operation influence energy use and indoor air quality in data centers and provides strategies to improve both design goals simultaneously.

As an initial step toward understanding data center air quality, measurements of particle concentrations were made at multiple operating northern California data centers. Ratios of measured particle concentrations in conventional data centers to the corresponding outside concentrations were significantly lower than those reported in the literature for office or residential buildings. Estimates using a material-balance model match well with empirical results, indicating that the dominant particle sources and losses - ventilation and filtration - have been characterized. Measurements taken at a data center using economizers show nearly an order of magnitude increase in particle concentration during economizer activity. However, even with the increase, the measured particle concentrations are still below concentration limits recommended in most industry standards.

The research proceeds by exploring the feasibility of using economizers in data centers while simultaneously controlling particle concentrations with high-quality air filtration. Physical and chemical properties of indoor and outdoor particles were analyzed at a data center using economizers and varying levels of air filtration efficiency. Results show that when improved filtration is used in combination with an economizer, the indoor/outdoor concentration ratios for most measured particle types were similar to the measurements when using conventional filtration without economizers. An energy analysis of the data center reveals that, even during the summer months, chiller savings from economizer use greatly outweigh the increase in fan power associated with improved filtration. These findings indicate that economizer use combined with improved filtration could significantly reduce data center energy demand while providing a level of protection from particles of outdoor origin similar to that observed with conventional design.

The emphasis of the dissertation then shifts to evaluate the energy benefits of economizer use in data centers under different design strategies. Economizer use with high ventilation rates is compared against an alternative, water-side economizer design that does not affect indoor particle concentrations. Building energy models are employed to estimate energy savings of both economizer designs for data centers in several climate zones in California. Results show that water-side economizers consistently provide less energy savings than air-side economizers, though the difference in savings varies by location. Model results also show that conventional limits on humidity levels in data centers can restrict the energy benefits of economizers.

The modeling efforts are then extended to estimate national data center energy use. Different size data centers are modeled to represent the national variation in efficiency and operation of associated mechanical equipment. Results indicate increased energy efficiency opportunities with larger data centers and highlight the importance of temperature setpoints in maximizing economizer efficiency. A bottom-up modeling approach is used to estimate current (2008) United States data center energy use at nearly 62-70 billion kWh annually. The model indicates that more about 65-70% of this energy demand can be avoided through energy efficient IT and cooling infrastructure design, equivalent to an annual energy efficiency resource of approximately 40-50 billion kWh available at a national level. Within the context of greenhouse gas emissions, benefits can be significantly increased by incorporating site location into energy-efficient design strategies.

The framework of this dissertation contributes to general building energy efficiency efforts by shifting the perspective of building design to address indoor and outdoor environmental impacts simultaneously, ensuring that one design goal does not eclipse the other. More specifically, the results presented here outline opportunities to temper the growing data center energy demand, so that IT can evolve into an energy efficient utility with the potential to facilitate a more sustainable expansion of goods and services.