The most widely accepted measure of data center efficiency is known as PUE or Power Usage Effectiveness has been developed by members of the Green Grid www.thegreengrid.org,  It is defined as.

Therefore to calculate PUE, the total energy used by the data center cooling infrastructure, UPS  input and ancillary equipment such as lighting, safety equipment etc must be collected.  Once this information is known, it is divided by the IT equipment load, which will be very close to the UPS output.  (There is some loss in transmission from the UPS to the IT equipment, typically less than 2%)

In the typical data center today, only about 50% of the energy is used by the IT equipment, thus the PUE in this instance is 2.0.  A data center running at a higher efficiency will have a lower PUE.  Conversely a less efficient site will have a higher PUE.

For some, having a lower number represent higher efficiency is considered counter intuitive.  A  derivative of PUE called DCiE or Data Center Infrastructure Efficiency has also been developed and adopted by the Green Grid.  This is simply calculated as.

This results in a higher number for higher efficiency.  A data center with a PUE of 2.0 will have a DCiE of 0.5 or 50%.  This means the site is 50% efficient.  If PUE were 1.6, a reasonable goal for many data centers, the DCiE would be 0.625 or 62.5% of the total energy going to IT equipment.

As of this writing, the average PUE for most reporting data centers is close to 1.9.  However accurate numbers are hard to quantify, since the metric is so new to the industry.  The EPA has set a goal to reduce average PUE numbers dramatically by 2011 depending on the type of infrastructure equipment.  There are a small number of data centers reporting PUE numbers better than 1.4 today.   The best reported PUE in North America as of this writing is 1.12 in the Pacific North West (a particularly good climate for data center efficiency).

There are a number of cautions when looking at PUE numbers published by others or collected for a specific data center.  PUE is a very dynamic number!  It will improve as the weather gets cooler such as evenings or cooler months.  Some data centers have very dynamic heat loads that change as computing demand changes.  PUE will often improve as computing loads go up and cooling systems run closer to capacity.  The most reliable PUE numbers are those that are an average of many months, hopefully a full year; though data center IT equipment profiles can change dramatically in a whole year.

There is another issue when comparing PUE numbers.  Accurate PUE numbers require proper sub metering of the entire data center only load.  This has proven difficult in multi use corporate facilities, less so in single purpose data center sites.   This is one reason a believable representative population is hard to quantify at this writing.  Colocation and IT intensive sites, by their nature, are motivated to improve efficiency since it goes to their bottom line.  They are also the most likely to publish their PUE number, particularly if it is good.  Corporate sites often don't know their real PUE, due to metering issues, and are more likely to have a worse PUE, simply because data centers are not their main corporate focus.  This situation is rapidly improving due to corporate sustainability awareness.

In summary PUE (and DCiE) is a good metric to understand measure and track.  It can vary greatly by climate and compute load as well as reporting methodology.  Focus on continually improving the PUE for a specific data center over time.  Don't get caught up comparing PUE to another site in a totally different climate.