Top 10 Things Data Centers Forget About PUE

Julius Neudorfer
In light of the reaction to Facebook's recent announcement of its new Prineville, Ore., data center, with its 1.07 power usage effectiveness (PUE) claim, which seems hyper-efficient (or perhaps just hyper), as well as many other very aggressive PUE claims, I thought we should all take a deep breath and see what is sometimes 'overlooked' in many data centers' PUE calculations:

Power Distribution Losses: Downstream of the UPS, these losses are sometimes ignored; typically they can represent three to five percent for 'copper' (I2R) losses, and sometimes up to five to seven percent if the PDUs contain a transformer. In fact, the Environmental Protection Agency (EPA) even found that a majority of the 120 original Energy Star for Data Centers program participants did not have the capability to measure the IT load 'at the plug.' As a result, the EPA is willing to accept the UPS output as the IT load, if more accurate IT load data is not available.  

Generator: The block heaters in backup generators require constant power to ensure the generators start quickly.

Support Areas: The power for required HVAC ventilation and lighting for related support areas such as electrical, battery and UPS rooms, as well as telecom/datacom entrance rooms.

Condenser or Chilled Water: If there is no dedicated condenser or chiller plant, and it is supplied from a third party or from within a mixed-use building, its energy value has been overlooked by the person making the PUE calculation. Also, some manufacturers of data center containers that use chilled water and expect conditioned power from an external UPS have aggressive PUE claims. In fact, in general, they should not be able claim a 'PUE' for their container only, unless they also provide all of the support infrastructure energy in the calculation, just like any other conventional site.
 
Fan Deck Power: The external DX condenser or glycol fluid dry coolers are many times excluded from measurements in existing sites, again particularly in mixed-use buildings.

Security Systems: Security and access control systems, as well as CCTV systems are often ignored, especially if they come from a dedicated power panel with separate backup power or are part of the main system in a mixed-use building.

Emergency Lighting Systems: They are (or should be) on their own dedicated backup power circuits and are often overlooked.

The Calendar: The time of year can lead to improper PUE calculation methodology. The original PUE calculation was based on a power reading, and many times it was taken as an instantaneous snapshot, under 'best case conditions' (i.e., coldest day of the year, running only on an economizer) and thus did not represent an actual 12-month average of energy usage. This is now a requirement and is part of the new Harmonizing PUE guidelines issued in February 2011 by The Green Grid, and adopted by the EPA, European Union and Japan.

Candor: If you are going to claim a very low PUE for its promotional value, then at least provide the basis of the calculation measurement and any related caveats or conditions (even if it is small print). It seems that there are almost always some losses that they think should have been included but were not because the marketing department said to ignore it since it would worsen its spectacular PUE claims.


Some of the items listed have significant implications, while others are small and may only have a minor impact on the final PUE calculation. However, when there are claims approaching a theoretically perfect 1.00 PUE, by implication they are laying claim to TOTAL infrastructure energy of less than 10 percent. That begs for closer scrutiny, or at least clarification of the conditions and basis for the calculation. 

The Green Grid has specific PUE protocols and if they are followed should provide a fair overall assessment of the energy efficiency of the data center's infrastructure. Toward that end, I thought that I would go right to the source for comment on my views on the 'sins of omission' for PUE claims.

In an email exchange with Mark Monroe, executive director of The Green Grid, I received this comment:

The Green Grid has worked to drive industry alignment of PUE through a consistent, repeatable measurement strategy and reporting processes. These efforts represent a significant step in advancing data center energy and resource efficiency on a global scale. Establishing a common understanding of efficiency metrics within the industry will help generate the necessary dialogue among data center owners and operators to improve efficiencies and reduce energy consumption within these facilities. 

The Bottom Line

In the short term, the race to be the greenest data center may produce some overly ambitious PUE claims. Some of these, were they to be fully examined, might prove to be exaggerated. Nonetheless, the long-term outlook indicates that there is now a common awareness and consensus in the industry that making data centers more energy efficient has many direct economic, as well as political, benefits to the owners and users, as well as the planet in general.

So next time someone claims that their data center has achieved a PUE of 0.9, ask them if they were part of the team of scientists that developed cold fusion.
 



Add Comment      Leave a comment on this blog post
Apr 28, 2011 8:04 PM Bill Tschudi Bill Tschudi  says:
I would add diesel fuel for standby generators to the list of often missing pieces of PUE. PUE is meant to account for all energy use on site and this includes the energy supplied by diesel for back up generators. Their testing and use energy use must be included in PUE. Reply
Apr 28, 2011 10:04 PM Julius Julius  says:
Good Point, I missed that one.. Thanks Bill... I guess I will need change the title to (Top 11 Things...) - Julius Reply
Apr 29, 2011 12:04 AM Andrew Stokes Andrew Stokes  says:
I know the Green Grid defines PUE as Total Load / IT Load, so in theory PUE is limited to 1.00 upwards. For a while, people tried to claim that any reuse of the power (e.g. to heat an aboretum) could be set off against the Total Load, thus allowing sub 1.00 PUE. This lead to the Green Grid's ERE definition and cautioning against calling this sub-1.00 PUE. Here's a thought though. The actual IT load is the excitation of electrons to move around in ways we recognize as data processing or data communication. This is actually a very tiny fraction of the power consumption of the server. In fact it's so small that we casually equate the "IT Load" directly with the "waste heat" from the servers. (Imagine, for example, judging the efficiency of a truck engine by the heat that the engine block generates, rather than the towing capacity, etc.) If you measured PUE like this (e.g. "true true PUE", to throw a shout-out to James Hamilton), then we are all running PUE's in the millions! Some work to do here. On the same vein ... what if you could use a thermoelectric device to recapture just a fraction of the waste heat and feed it back directly as electricity to the servers. In this case, given you would then literally be running more IT load (IT waste heat load - heh!) than facility gross power, I argue that this would indeed be a true reason to break the sub-1.00 PUE barrier. Reply
Apr 29, 2011 6:04 PM Phil Hughes Phil Hughes  says:
These are a great way of hiding true PUE. Even at idle fans take 7-10% 0f IT load bur can crank up to 40% or more when the inlet air gets hot. Suspending disbelief for a moment and an agreeing to Facebook's claimed PUE, even on a cold day the "true true" PUE = (1.07+.07)/(1-.07) = 1.23 and on a really hot day PUE = (1.07+.4)/(1-.4)= 2.45 Reply
Apr 29, 2011 8:04 PM Tony Khoury Tony Khoury  says:
Last time I looked, the PUE calculation is supposed to include the Kw hr usage taken from the utility bills for the previous 12 months for the data centre and so when PUE analysis is done correctly all the electrical losses within the facility should come into play. Facebook has been quoted as saying that their electrical system only incurs a 7% loss (which is typically a good number) and when you do the numbers using only this loss, without taking into account the electrical usage of the cooling, lighting etc means they are running at a PUE of 1.075 so something seems wrong with their stated PUE of 1.07 Reply
May 1, 2011 4:05 AM Tim McCain Tim McCain  says:
Take a look at this.. Reply
May 1, 2011 7:05 AM Ronald Ronald  says:
Imho PUE measurement is dead simple: Measure the incoming power to the utility and measure the IT-load (preferably as close to the IT equipment as possible). Measure these values constantly. Devide the incoming power by the sum of all IT-loads and you have your actual PUE. All other ways will lead to a PUE value that is corrupted by the emission of one or more of the 11 (?) things above. And yes... a PUE is not a constant number. It varies permanently. Datacenters that don't measure IT-loads should not be allowed to publish PUE numbers. We all know that theoratical PUE have no meaning at all. Measuring it is knowing it ! Reply
May 2, 2011 3:14 PM Scott Payton Scott Payton  says: in response to Tony Khoury
That is easier to do if the building is a built for purpose data center only. But much harder for facilities that exist within sites, which are still legitimation Data Centers (in some cases). Without that, you need some way to measure all the incoming power. That gets tricky when the DC lights are still part of the base building, or when sharing UPS room with other facilities that aren't part of the DC. This gets really tricky then. Reply
May 3, 2011 5:05 PM Michael McKenna Michael McKenna  says:
In a conventional data center UPS losses account for anywhere between 3-10% losses. Also fan power for conventional CRAH units are about 10%. Even if you account for the fact that a full air economizer system has little or no air coil resistance, the filters and intake louvers do account for something. Reply
May 9, 2011 1:43 PM Todd Owen Todd Owen  says: in response to Phil Hughes
@Phil: The idle fan numbers you cited are consistent with 1U servers, sometimes stacked in series. The Facebook data center design is different; the server fan energy is much lower, primarily due to pressurizing the cold aisle and the 1.5U server design. @Sub 1.00 PUE: See Green Revolution Cooling (grcooling.com) for some real sub 1.00 PUE solutions. You're basically subtracting 10+% of the IT load from the server since the fans are removed. I would love to see servers output their fan power over SNMP so it could be monitored--Dell, HP, IBM ? Reply
May 24, 2011 2:05 PM Ralph Ralph  says:
This is serious and REAL and the US government has been working on it for 10 years. Please take 15 min and explore the link provided Andrea Rossi has given three demonstrations so far including with professors from Bologna University and the Swedish skeptics society and the Chairman of the Swedish Physics Union. This is an directory of Rossi efforts http://peswiki.com/index.php/Directory:Andrea_A._Rossi_Cold_Fusion_Generator. This is a link to the LENR site where detailed information about cold fusion efforts is available. www.lenr-canr.org/News... The US Naval Research lab has been working on this with positive results for over 10 years and has confirmed it existence. Yet the major scientific magazines refuse to touch this issue since it was purportedly discredited by some researchers and an institution that stood to lose 10s of millions in funding per year in hot fusion. Government funded hot fusion systems have never produced surplus energy after years of research billions invested. Rossi has announced a 1MW Cold Fusion facility to be opened in Greece this Oct. Still top line periodicals have yet to publish even one article. This will change the economics of the world lifting many people out of poverty and it will also threaten many vested interests. http://pesn.com/2011/05/17/9501827_Ampenergo_Amps_Up_Rossis_Energy_Catalyzer_in_America/ "..Ampenergo was founded by Karl Norwood, Richard Noceti, Robert Gentile and Craig Cassarino. It is important to note that Robert Gentile was the Assistant Secretary of Energy for Fossil Energy at the U.S. Department of Energy (DOE) during the early 1990's. This helps confirm Rossi's claim that tests of the E-Cat have been observed by the U.S. Department of Defense and the DOE. It is very likely that at least certain individuals in the DOD and DOE are aware and interested in the Energy Catalyzer. However, their silence is deafening. It is unknown if any military or secret government research is taking place, but there are unsubstantiated rumors floating around the internet of the US Navy using a nickel-hydrogen cold fusion reactor to power a submarine. Although the rumor is not likely to be true, if they have known about the technology for a couple of years, it is possible testing is taking place. Trillions of dollars go missing from the DOD budget on a regular basis, and the money is obviously being spent on something..." Reply

Post a comment

 

 

 

 


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

 

null
null

 

Subscribe to our Newsletters

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