The Data Center Professionals Network
The Global Database of Data Center Industry Expertise
What's your PUE? ...and how can you prove it?
I recently watched a promotional video from data center company BladeRoom.
They claim a PUE of 1.15 which, according to their video, "is the lowest proven PUE value that they are aware of anywhere in the world today". Now, I don't know the date the video was made - but it looks fairly recent (although on their website they state that their London Datacentre has a PUE of 1.13).
This raises two questions:
1. What's your PUE? ....and how can you prove it?
2. What's the lowest PUE you know of?
Tags: PUE
Replies to This Post
-
Permalink Reply by Kim Ream on -
Chatsworth Products has a PUE calculator that helps you determine your potential PUE when using passive cooling cabinets with KyotoCooling, water-side economizers or evaporative air-side economizers. www.chatsworth.com/pue-calculator
-
Permalink Reply by Tony Fiori on -
I have looked at the Square D Powelogic ION Meters from Schnieder Electric and these meters will ablsolutely give you your PUE. Its kind of nice to be able to walk a person who cares up to a panel and show them on the meter the PUE. I havent witnessed anything under a 1.4 myself.
-
Permalink Reply by Tez Gale on
-
The figures BladeRoom are claiming are possible. However I cannot see any mention about which period they are run over. Having built a very efficient data centre ourselves using the same sort of techniques, we know that the PUE can vary over the whole year. From what I can see, BladeRoom have run tests for a few weeks at various settings and claimed their PUE. To get a true PUE of the data centre they need to run the centre for 12 months and provide the 12 month average.
As I mentioned we have a centre that is highly efficient. It has now been running for 20 months and so we can now provide real data. We have a trailing twelve-month (TTM) figure that is calculated using total kWh readings taken at certain points in the year. The calculations use exact figures and not averages. We have managed to run the centre at a PUE of 1.153 for the last TTM. This is a centre based in London and is not at full capacity. See http://www.keysource.co.uk/successstories/index.aspx?id=813 for more details.
The question of the lowest PUE should never be asked. PUE is dependent on so many factors - location, redundancy options, type of IT equipment, Tier level etc. All companies should be striving for an efficient centre and trying to reduce the power coming into the data centre to the lowest possible for the level of resilience required. PUE can also be manipulated. It is easy to get a lower PUE by running a cpuburn on all processors. I would rather see the switching off of equipment if it is idle even if this raises a PUE. The key is to lower the power coming in.
PUE is great; it is a good indication that a centre is efficient, but it is NOT something that should be compared between data centres. Comparing a Tier 1 centre in the Sahara desert, to a Tier 4 in Iceland is like comparing apples with a giraffe.
What is needed by all data centre operators is to review how their data centre is set up. Questions like these below need to be asked:
Do I need this level of redundancy?
Can I virtulise to reduce servers (not always possible - we don't do it)?
Am I separating hot and cold air?
Am I making use of the outside air and using free cooling?
Why am I running all this equipment? Do I have things that I can turn off?
Am I provisioning the data centre for the load I have or for a possible load in the future?
An apology if this has started to turn into a rant. I'm just quite passionate about this.
-
Permalink Reply by DCPNet Admin on -
All good points Tez. Doesn't come across as a rant, but can definitely tell your passionate about it :-)
Question for you and anyone else reading this:
From your experience in the industry, how common is it to see PUE figures that state that they have been measured over a 12 month average?
-
Permalink Reply by Tez Gale on
-
The only people I have seen publish this is Google. They measure all their new 5MW+ sites with this. They don't measure their smaller or older sites. But hey its a start.
-
Permalink Reply by Mohammed Shehata on -
I personally have a problem with PUE as a benchmark, and I agree with the analogy that Tez gave of apples and giraffes, but I can go one further where the same DC, same tier, same equipment (exclude free-cooling) and built in a 'clean' energy location vs. say China or the US, then whilst the PUE is the same, the carbon footprint is very different. Surely energy efficiency is not just about potential operational cost savings but also about environmental impact?
I was very disappointed when the EPA abandoned their work on tracking PUE back to the source and decided to ignore it, as I felt that was the right direction (albeit they were only looking at losses on the generation and distribution side rather than the generation source itself).
How low can you go with PUE is dependant on when and how you measure it, and I think Tez covered most of this and in particular the point about the 12 month average. I do however believe that equipment types should be looked at in more detail as well as % capacity utilised for that PUE average figure along with the Tier for benchmarking to be considered. Most equipment (e.g. UPSs will have a variable efficiency characteristic based on load, and if 12 or 6 pulse rectifier technologies are used instead of IGBT or CSTBT then the tier class and as such % capacity will make a huge difference to the PUE no matter how good your cooling COP is).
From my perspective, I would like the PUE to be published with the following details:
1. Design PUE
2. Measured PUE during T&C (during full load heat load tests)
3. Twelve Month Average PUE (this will take into account the seasonal changes as well as varying load profile for the daily, weekly and monthly usage)
4. Design Tier class
5. Graphical representation/trend of the load profile for the 12 month period
6. Type of business that the DC is supporting
7. Type of UPS and cooling systems used
-
Permalink Reply by Michiel van Blommestein on -
I have to agree with Tez and Mohammed. The main problem is that PUE is not designed as a metric to compare different facilities with each other; PUE is designed to provide a benchmark for the operational efficiency of a single facility, which you can use to get some idea where the efficiency problems are lying.
It is worrying that PUE is more and more used for comparing, and less and less as a benchmark. It risks becoming a slogan, as the whole 'pH-neutral' thing in commercials for dish washing soap is. PUE becomes meaningless.
I am aware that there is some kind of standardisation going on, but I think that any claim with PUE should be taken with a bag of salt.
BTW: 1.15 is absolutely ridiculous anyway.
-
Permalink Reply by Tez Gale on
-
1.15 is absolutely ridiculous? Is that in a good way, or you don't believe it? Either way, yes it should be taken with a pinch of salt (even though the figures can be proved) as you'll see below.
I agree with you about benchmarks. I know there are several attempts at creating better benchmarks that compare actual work performed compared with energy going into the centre. No one yet has been able to come up with a good formula. The big thing is no one can state what "actual work" means. There is no way to work out a metric that is meaningful without this.
My view on this is that by the time you try to work out how productive you are, you have wasted so much energy calculating the efficiency that you have used up your savings in the cost of getting the measurement. The whole exercise becomes rather futile.
PUE is therefore the best measurement we have so far. It should never be used to compare sites though. However we have used it to compare the data centre with itself. If we notice that the PUE has gone up we know there is an issue. This helped particularly in the early days of the centre to balance the cooling system to gain optimum power gains.
As an example, we found our PUE went up dramatically between 16-19C ambient outside temperature. This was traced back to fans ramping up to full speed before the adiabatic cooling kicked in. By kicking the adiabatic cooling in 2 degrees lower we saves about 50-60% of the cooling costs for that temperature range.
So in conclusion PUE is great for comparing a data centre to itself, but not for comparing to other centres.
Changing tact a bit. I think that signing up for the European Code of Conduct and following their best practice principles will give anyone a more efficient centre.
-
Permalink Reply by Michiel van Blommestein on
-
Tez - Thanks for the background! To be honest, I did not trust the 1.15 figure because it is well below what is generally accepted as being the 'theoretical minimum', but hey: it is all just theory anyway, right?
I do agree with you that, as a formula. PUE is great for one, simple reason: it is very straightforward. That makes it superior as a year-long benchmark to all other measurements out there.
-
Permalink Reply by Engel Nobbe on
-
I have been managing a datacenter which used (do not laugh) DX units. It achieved 1.52 over time, but I have to add that all kit was new, ie. 2 years and younger. I'm no longer involved with the specific company so I cannot tell you how it is performing right now.
EUE's (which I prefer to translate to 'average PUE') of 1.5 should be quite possible. EUE's of 1.3 are possible as well, but hard to achieve. Most EUE claims are somewhat, euh, creative, so I can understand Bladeroom's statement about 'proven' EUE.
PUE in itself means nothing though, unless it's average PUE.
-
Permalink Reply by Paul Poetsma on -
Just ran into a new DC with a PUE of 1.12 as average over May 2010. It was a very cold month where their free-cooling capable data center handled an IT load growth of over 30% easily. Other months show 1.2 up to 1.5 and it fully depends on the weather. I look forward to see their yearly numbers.
Welcome to
The Data Center Professionals Network
© 2012 Created by DCPNet Admin.
