@misc{29655,
  author = {Mohan Ganeshalingam and Arman Shehabi and Louis-Benoit Desroches},
  title = {Shining a Light on Small Data Centers in the U.S.},
  abstract = {<p>Large data centers are well known for their high-energy intensity and have made dramatic efficiency improvements over the past decade. Small closet and room data centers have received much less attention, yet constitute a significant fraction of the total number of servers in the United States. The often makeshift, ad hoc nature of small data centers often results in little attention paid to energy efficiency and inadequate cooling equipment. The small physical footprint of these data centers, typically embedded within a larger building, makes it difficult to identify and target for efficiency measures. These conditions make small data centers notoriously inefficient relative to their larger counterparts. In this report, we present an analysis of small and midsize data centers in the US, drawing from surveys of commercial building stock. We find that servers in small data centers make up approximately 40% of installed server stock, with the vast majority of sites utilizing only 1-2 servers. We identify industries where small data centers are most prevalent, finding that the highest saturations are in medical, retail, office, and education sectors. Small data centers typically lack dedicated cooling equipment, often relying on building air conditioning and ventilation equipment for cooling. We further find that the type of cooling equipment used is highly correlated with the number of operational server racks, with less efficient cooling options used with fewer racks. We develop geospatial maps of small and midsize data centers to visually identify regions of high server concentration and calculate associated CO<sub>2</sub> emissions. Small data centers consume 13 billion kWh of energy annually, emitting 7 million metric tons (MMT) of carbon dioxide–the equivalent emissions of approximately 2.3 coal-fired plants. We discuss efficiency measures that could be implemented and estimate potential energy and CO<sub>2</sub> savings.</p>},
  year = {2017},
  month = {06/2017},
}