PUE: Cold Con-Fusion in the Data Center

Julius Neudorfer

PUE, DCiE and EUE and many more acronyms are all presumably standardized metrics used to measure and rank the energy efficiency of our data center infrastructure. Yet, these terms are used and misused, calculated and miscalculated on a daily basis by many people. There are those in the industry that just want to measure their own data center energy use, as a starting pointing, with the ultimate goal of trying to improve their data center energy efficiency.  

Yet, others simply want to have green 'bragging rights' by proclaiming a very low or 'the lowest' PUE.   Of course, to clarify, PUE is 'Power Usage Effectiveness,' and its mathematical reciprocal DCiE is 'Data Center Infrastructure Efficiency' are terms created by The Green Grid in 2007.

It is still unclear to me why the DCiE, which is a direct reciprocal of PUE, continues to exist. But what I find disheartening is that many of those professing to measure their PUE don't really understand how to correctly calculate it (where and how to measure) or don't have the correct measurement equipment in place to properly measure the two 'basic' numbers used in the PUE formula.

The PUE is defined as follows:

PUE = Total facility power / IT equipment power

The most common problem is that many people in the IT world (and even some in the facilities world) do not fully understand the difference between the words 'energy' and 'power,' and the terms KVA and KW or KWH. Sometimes they even use them interchangeably.

Just a brief clarification of the above terms:

KVA is simply (volt x amps) / 1000-also called 'apparent power' 
Example (single phase) 208V @ 200A = (208 x 200) / 1000 = 41.6 KVA

KW is (volt x amps) / 1000 x power factor-also called 'true power'
Example: (single phase) 208V @ 200A = (208 x 200) / 1000 x 0.9 PF = 37.44 KW

Power is express as KW.
Energy is power x time, expressed as KWH.
Example: 1 KW for 24 hours = 24 KWH.

The Myth: You cannot use apparent power (KVA) to calculate PUE. You must use KW (true power). Your metering system must be able measure KW to provide a meaningful PUE. (preferably it should also measure KWH)

I believe that the introduction of the PUE metric by The Green Grid has done a lot of good by focusing attention on data center energy use and more importantly, how to improve it.  One of the inherent flaws of the 'Basic PUE' metric is that by definition it is calculated based on 'power" (KW), which is an instantaneous measurement. It is not an 'energy' measurement based on a period of time, i.e. a month or year, which would be expressed in KWH.

However, the underlying issue with PUE, even when properly done, is that it is a snapshot of the power at that moment. It can be taken at any time (i.e. in the middle winter of the coldest night - running on 100 percent economizer-based cooling, with no chiller/compressors running).  To represent a realistic efficiency model, it should be based on a 12-month continuous average of the true power (KW). The Green Grid refers to this as 'Level 3-Advanced - Continuous Monitoring.'  In my opinion, there should simply be an 'annual PUE' plainly calculated as the total of 12 months of IT energy (KWH) / total facility energy (KWH).

Even the EPA's pending Energy Star for Data Centers program, which examined more than 100 sites over a 12-month period, found that most of those data centers did not have the key points properly metered to measure true power or energy.  In fact, they ultimately settled for simply accepting the 'output of the UPS' to represent the IT energy. At the very least, this ignores the downstream distribution losses and any step-down transformer losses.  Worst case, some data centers power the blowers of the CRAH units from the UPS, thus including them in the IT load, instead of the mechanical cooling (total facility energy). 

The Hype. I have seen many claims by data center operators and equipment vendors, and even some articles, proudly proclaiming a very 'low' PUE. Yet in many cases, they did not properly implement The Green Grid's prescribed measurement points to correctly capture valid KW information. I have even seen KVA used to 'calculate' the PUE in articles. Beware of PUE results that approach 1.0; the average still seems to be 2.0, however 1.8 or better can achieved without extraordinary measures. Nonetheless, if you have calculated your own PUE as 0.95, congratulations your data center has achieved 'cold fusion.'

So here is my Truth or Dare: If you are proclaiming how efficient your data center (or product) is by citing your 'PUE,' my dare is for you to properly document  how, where and when you calculated your PUE.

All that being said here is my own caveat:  If the only information you have available is from the UPS output and total facilities KVA, by all means use it to create your efficiency ratio baseline and then monitor it for any changes as you make improvements to your data center. This will still provide you with a good relative indicator of how much you have improved your power efficiency, which is the ultimate goal.

Add Comment      Leave a comment on this blog post
Feb 4, 2010 10:02 AM Kieran Storey Kieran Storey  says:
Julius, Thank you for this article, as you say using KW rating at an instant in time is completely incorrect especially if you consider the common plant arrangements that are used for data centre applications (air side economizers & water side economizers), also the variation in chiller efficiency over the course of the year for this type of application is far greater than a standard commerical type AC system due to its 24/7 season operation. Kieran Reply
Apr 30, 2010 11:04 AM Anonymous Anonymous  says:
Julius, great article and all very good points. PUE has been transformational in getting our industry motivated about energy efficiency but it is open to manipulation. In addition to your points we should also be aware that IT equipment with inefficient fans and power supplies can have the effect improving PUE but not delivering efficiency. A low PUE doesn't always deliver lower energy bills. Perhaps there is an argument for a 'modified' PUE where fans and power supplies are moved to the other side of the PUE calculation ? Ciar�n Reply
Feb 8, 2012 3:02 PM lmele lmele  says:
Julius, with reference to your opionion ("there should simply be an �annual PUE� plainly calculated as the total of 12 months of IT energy (KWH) / total facility energy (KWH)"), how could you calculate these values? Let's consider, for instance, that I take instantaneous total power and IT power, once a month. At the end of the year, how do I get the final values fo the total and IT power? By calculating an average? Thanks. Lorenzo Reply

Post a comment





(Maximum characters: 1200). You have 1200 characters left.



Subscribe to our Newsletters

Sign up now and get the best business technology insights direct to your inbox.