PUE - How to Lie With Statistics

EVERYWHERE you turn today, metrics leap out at you for energy efficiency. What the statistics don't offer is a normalization of the data when comparing various facilities in various areas. Simply stated, a data center sucking in cold air off the North Atlantic or Baltic Sea and can cool the data center without some form of consistent, mechanical temperature conversion plant will certainly use less non-IT power that a facility located in lower Manhattan or Miami.

Statistics are only relevant when they can be related to a wider data pool. So, to comment that you have 1.20 PUE may be super in Silicon Valley, but not so good in Amsterdam.

Similarly, generating PUE data on days advantaged for the data, like the coldest day of winter with 100% outside air may look good on paper but does not speak to the true nature of the metric. Here are a few positions we offer our clients when we discuss PUE and making it relevant and defensible to your firm:

• Make your PUE an average and annualized reading, not a "best case" figure.
• Test your systems and measure your PUE on the "design day" for the mechanical system, or the hottest and most humid day of the year. In short, capture the "worst case" basis of the mechanical system. It also serves an an excellent benchmark of how much cooling capacity you have left, or don't have.
• Back up your PUE or DCiE calcs with the work of Greene Grid and the accepted DC standards committees.