Here we go again! This time it’s Sun Microsystems.

I have been asked by a client to comment on PUE calculations by Sun. I don’t have the presentation and this could have been an example and not meant to portray a real data center. So I’ll give Sun some latitude.

Sun calaculate’s the PUE for a data center identified as Sxxxxx as follows:

Chiller plant – 126kW (vs. an industry average of 399kW). WOW! 798kW of IT 39kW losses 20kW requires a size chiller (minimum) of 243 tons, but this does not include removal of heat absorbed thru the building shell, people or latent cooling. These loads would probably bump this number up by another 20 to 50Tons. We will ignore this as we can make our point with the most conservative of numbers.

If we assume that the 126kW figure include pumps and cooling towers, then we need to do some guessing. For sake of argument, we will guess low (We want to be fair to Sun). So lets say a 10Hp Pump 10 Hp fan, roughly 20kW. So the chiller efficiency would be 100kW/243Tons or 0.41 kW/Ton. I don’t think that you can get a chiller any more efficient than 0.47kW/Ton. If we accounted for all of the load s, the chiller would likely be closer to 300 Tons requiring a chiller of 0.33kW/Ton.

For the calculation of electrical losses, to be correct, this figure would need to include the feeder losses, UPS losses, PDU/Transformer Losses and branch circuit losses. Wow! For kicks, I just picked a size and looked up the numbers for a Square D 150kVA “energy efficient transformer” - &#xIR;=3.78. I also looked up a Powerware 9395 and the efficiency is specified at >94%, Liebert 610 -1000kVA system 93% at full load. Now, it’s a bit unreasonable to assume that their is no redundancy built into the system. Redundancy would of course reduce the overall efficiency of the system. A reasonable figure for feeders losses is probably something close to 3%. So somehow Sun has managed 3.86% total (vs. 10% industry average). Yes there are other manufacturers and models of equipment and viable approaches to reducing the power losses but 3.86% is quite a stretch.

Lighting! Yes we all need lighting. They show an industry average of 3% and that they have reduce it to 2%. Ok I give. I don’t know a single engineer that doesn’t specify the most efficient lighting fixtures and reasonable lighting levels. So bring your flashlight when you visit.

In short I don’t think Sun could achieve the number that they are claiming. Its pretty easy to be a star when nobody is checking the numbers!

Sun has saved the owner some $400,000 annually. To get this they simple take the kW savings x 7 x 24 x 52 x $0.08/kWh. This is a gross over simplification and simply erroneous.

Efficient design starts with appropriate system selection and selection of efficient equipment. But it also includes (and I would be quite disappointed with my engineers they missed it) energy savings techniques such as water side and/or air side economizers, VFD drives on the tower and pumps, occupancy sensors on the lighting, etc. They also failed to consider that data center loads normally vary as much as 10% between heavy demand and slack demand and with newer technologies such as speed throttling offered by AMD and Intel, we expect that this variation will get larger (i.e. reducing thier savings estimate).

Sun either missed this or didn’t quite know how to figure it into their energy savings calculation. Hint: It takes a bit of modeling. In any case, the savings estimate is overstated. Hopefully, it's only their presentation thats wrong.

My point isn’t to bash Sun. While the numbers appear to be made up, the degree of error is certainly much less then Gartner’s (See previous Blog)

My point is that there is a tremendous amount of misinformation that is being promoted by members of this industry. Some of it is out of ignorance or perhaps adherence to old wives tales. Some of it comes from applying good, factual information to the wrong situation. Some might be legitimate mistakes. Some of it is bold face lies to sell more equipment and services.

Now, so that you don’t think me a Pollyanna, I know that everyone embellishes the facts, including me, and different engineers might look at a given set of conditions/facts and legitimately form different opinions and make different recommendations. Many times there are multiple ways to get to the end task.

I don’t disagree with the concept of PUE but lets use real numbers and correct calculations.

Our industry is seeing great change, it’s in transition and there is a lot of confusion. Buyer beware! If it sounds to good to be true, it probably is!

PS: A consultant should be someone who works on your behalf, someone you can TRUST! If you enjoy reading, try “Clients for Life” by Jagdish Seth and Andrew Sobel

Comments (4)

Subscribe to this comment's feed

...
written by raid , September 19, 2008

I have been wondering how companies can claim PUE figures that are somewhat hard to believe for some time. I think I now understand how this could done.

I have been assuming that the PUE figures are based in premium sites. But lets for argument sake, assume that this is not a premium site but a test facility. The site has more then 90% on its load on raw power (non-UPS). Also let’s assume that being a test facility the site does not bother maintaining the temperature and humidity of the facility to ASHRAE class 1 standard. This will allow for very large air side economy cycles so that the air-conditioning plant is only used in the hottest parts of the day.

If you use the methods above I feel sure that you could obtain a very good PUE.

report abuse

vote down

vote up

Votes: +0

...
written by kenbaudry , September 20, 2008

What you have hit on is that trying to compare the PUE of a Tier I facility to a Tier IV facility is like comparing an apple to and an orange. This is just one of the fundamental problems with PUE calcs.

From a high level view PUE is based on adding up all of the “nameplate ratings” on the facility equipment (air-conditioning, lighting, pumps, etc.). It has nothing to do with how the equipment is operated.

There are a couple of MAJOR problems with this concept. First, some choose not to include certain equipment. Why? If I can come up with a good rationalization, why not? There are no “PUE Police” to say differently.

Second, the end result of PUE tells us practically nothing about energy usage! If you and I both have the same brand and model car, then who will spend more on gas over the course of a year? Will we get the same gas mileage? Even if we have different but comparable cars (Mazda, Accord, Camry), we can’t answer these questions without knowing the driving habits and how much each of us drive every year.

It’s easy to see that optimizing facility operations thru air and water side economizer, proper placement of floor tiles, use of closure panels in cabinets to block bypass air thru the floor, cold/hot aisle orientation of equipment, proper set point for temperature/relative humidity, etc. play a very significant role in determining our energy cost.

Finally, PUE does not consider whether the facility meets the business requirement of the user. This is the most basic of all measures; Get it Right! Over building, over specifying performance levels results in higher first cost, higher energy usage then necessary regardless of how efficient the equipment and systems are and regardless of how well your operate it. Back to the car analogy driving the most efficient bus with only two passengers will never be as cost effective as driving a compact car, and making twenty trips with a compact car to transport a busload of passengers will never be as efficient as a bus. Failure to match the facility design to the need makes any measure of energy effectiveness worthless.


The bottom line is always money and it’s important to have a measure of how well we are doing, to know whether we are operating more efficient or less efficient that our peers in the industry.

What is needed is 1) a better appraoch to matching design to needs, 2 a predictor of energy consumption that can be used when designing new facilities and 3) an ongoing program that MEASURES our performance. PUE for the most part, fails to achieve either of these objectives.

Anything short of this is a waste of time, energy and effort (ungreen).

report abuse

vote down

vote up

Votes: +1

...
written by ANECLTD , November 12, 2008

We have addressed many of these issues with a purposely designed system and can collect data in real-time from all devices and the support infrastructure. We intend to address the issues of providing processed data to deliver efficiency values in the next few weeks. Our product is called ONCALL and we based in the UK. Please contact us directly if you want to have a look at this application. We may have just what you need.

Chris J Knight (skype:cknight2)
Founder: ANEC Limited ( This e-mail address is being protected from spam bots, you need JavaScript enabled to view it )

report abuse

vote down

vote up

Votes: +0

...
written by Ariel , May 14, 2009

Agree with Kendaudry that PUE is not a perfect solution. But at least it is one of the most effcient practical one. At least you can have an estimation and compare with industry to identify the problem.

And there are always some free tools online, like: PUE and DCiE Efficiency Savings Calculatorshttp://www.42u.com/measurement/pue-dcie.htm

I will do the calculation from time to time anyway, just to have a picture in mind. Better than nothing.

report abuse

vote down

vote up

Votes: +1