There’s a new data centre efficiency measure. Peter Judge does the maths but remains unconvinced
Once again there is discussion about “replacing” PUE, the venerable and effective standard for measuring data centre energy efficiency. This time, it’s coming from a European standards body led by telecoms operators and the results could be a problem.
PUE has succeeded because it is simple and workable. The new DCEM, which has come from ETSI (The European Telecommunications Standards Institute) makes things more complicated.
A standard to bury PUE?
ETSI’s DCEM (Data Centre Energy Management) standard was published by ETSI in June after two years’ discussion within a French IT managers’ group called CRIP. The press release and the full standard (ETSI ES 205 200-2-1) are online but, apart from press release feeds, the announcement was only picked up by French news outlets.
There are different takes on this. Silicon.fr thinks this is a standard which will “bury PUE”. Other publications say similar things, mostly based on quotes from Dominiqe Roche, the head of sustainability at French operator Orange, and clearly a prime mover in this effort.
Here is one fairly typical quote from Roche: “PUE is dead, because it is not usable at the operational level. This is certainly the view of Europe.”
GreenIT.fr sees it differenlty, calling DECM “normalised PUE” and “the equivalent of PUE as defined by the Green Grid”.
The truth is that the new ETSI proposal does include PUE (more or less), though it doesn’t refer to it by name or mention the Green Grid. But it also includes other factors such as the renewable energy consumption, energy re-use, and the size of the data centre.
Skip the maths if you must
Before we look at ETSI’s equation here is a reminder of PUE (power usage effectiveness) as defined by the Green Grid.
Now, here is the DCEM equation from ETSI.
KPIEM = KPIEC x KPITE x (1- KPIREN x WREN)) x (1− (KPIREUSE ×WREUSE))
The following text contains plenty of equations. If you don’t like this, skip to the next cross-head. If you want more, then also read the full standard. Maths, like words, shouldn’t be taken out of context, and you shouldn’t assume I have understood it completely. But here is where I have got to.
KPIEM is the thing we are looking for, a measurement (a key performance indicator) of efficiency.
It looks complicated, and I suppose it is – but we can see roughly what is going on here. KPIEM is essentially based around a highly modified version of PUE, which ETSI calls KPITE , the Task Efficiency measure.
You can see this PUE-like thing, KPITE, at the centre of the equation, but it is multiplied by three other things. On the right are two factors which reward data centres for consuming renewable energy and for re-using waste heat.
The reward factor for using renewable energy is this one: (1- KPIREN x WREN)). The KPIREN factor is the fraction of renewable energy used, and WREN is a weighting factor which ETSI chooses to set as 0.8.
So essentially, if you switch entirely to renewables, like Apple, your overall score goes down by 80 percent.
The reward factor for re-using waste heat energy looks like this: (1− (KPIREUSE ×WREUSE)). The KPIREUSE factor measures the amount of waste heat energy that is re-used, and WREUSE is a weighting factor, which ETSI sets as 1.
This is a more complicated deal than the renewables modifier, but effectively reduces the overall score if waste heat energy is re-used. It can reduce the Task Efficiency measure to less than one, if the site has a very low PUE and resuses a lot of wasted heat.
Finally, the metric is multiplied by KPIEC – a measure of your total energy consumption in kiloWatt hours. This means that the overall figure is measured in kWh, and will be bigger for bigger data centres.
Data centre people I’ve spoken to are scratching their heads over the reason for this last bit. It seems to mean that the score for any data centre can’t be easily compared with any other. But when we’ve understood why it is there, we’ll let you know.
Classifying data centres like fridges?
So far, so complex. But ETSI understands most people don’t want that complexity, and attempts to create something usable. It boils the figures down.
It classifies data centres into four sizes (S, M, L and XL) asccording to how much energy they use. Small data centres use less than 1GWh of energy per year, Medium ones are between 1GWh and 4GWh, Large ones consume from 4GWh to 20GWh, and XL data centres are bigger than that.
Within these groups, data centres are grouped by “Data Centre Performance” into ten grades, from A to I, according to how efficient they are. This means consumers should be able to buy data centre services as they would buy a fridge or a washing machine, perhaps looking for a class A data centre service in the size they require. Large organisations can mark their own work from A to I in order to sell it.
But “a data centre isn’t a fridge”, says Liam Newcombe of Romonet in a blog. Among other things, this classification cuts across other ones such as the Uptime Institute ranking of data centres on reliability, based on Tiers. I also have not yet completely understood how DCP relates to the KPIs in the ETSI document – it is possible that DCP is simply equal to KPIEM without the KPIEC factor.
In one sense, data centre operators don’t need to worry much about DCEM, because PUE is already on its way to being enshrined in a standard from ISO, which is at the top of the tree of standards bodies. ETSI contributes to ISO, but would have to harmonise anything with ISO’s version of PUE version before it goes global.
But in another sense, this could be serious. Le Monde Informatique reports that 30 data centres have already assessed their DCEM figure, and shared results in an anonymised database. The same article hints that the standard could be “imposed” on suppliers in some sort of certification.
It’s clear that the people behind the ETSI work have good aims. “This KPI has been designed to match and fulfill the energy-efficiency commitment signed by the EU to apply the Kyoto environment protocol,” says French journalist (and NetMediaEurope directo) Pierre Mangin, adding that PUE does not include data centre life cycles or their exploitation of re-use and renewable sources.
But there is a difference between including these factors, and including them well. It’s not clear to me that there is a need to crunch the numbers into one figure, as this loses the detail and assumes all data centres can be treated the same.
We recently heard that the European Union wants to regulate data centre efficiency. There is a danger that it may pick up an over-complex standard. This could create perverse incentives and unforeseen consequences, and could produce a wasteful parasitic industry, making and marketing DCEM measurements simply to achieve certification.
A version of this article appeared on Green Data Centre News