Datacentre power metric meets with resistance

A low Power Usage Effectiveness rating is not necessarily a sign of 'going green'

Power usage effectiveness (PUE) is a powerful term because it gives a ballpark reading of the efficiency with which a datacentre consumes electricity. It is also a young term. It began to become noticed and adopted on a wide scale when The Green Grid, a global consortium of IT companies and professionals, began to publicise and detail the metric and encourage its adoption in 2007.

It was born because of the price of electricity. In 2007 US datacentres consumed 1.5 per cent (60 billion kilowatt-hours) of the country's total electricity consumption, according to a report by the Environmental Protection Agency, which predicted that figure to almost double again to 100 billion kWh in five years' time.

Because PUE indicates the power efficiency of a datacentre, it has become ever more useful as the industry, and its electricity bills, have grown.

You can calculate PUE on the back of a napkin. You divide the total facility power (TFP) by the IT equipment power (ITP). The TFP is the total power which goes into the entire datacentre facility, taken from the utility meter. The ITP is the power which goes into the things which "manage, store or route data within the raised floor space", according to The Green Grid definition. In other words, the electricity used by the ITP is the only part of the datacentre that makes money; everything else is just a support network.

If my TFP is 1,000 but the actual cost of my ITP is 400, then my PUE is worked out by dividing 1,000 by 400. This gives a reading of 2.5 which means that, for every unit spent on the datacentre ITP, 1.5 units are being spent on the infrastructure to keep the ITP centre going.

In the hands of a knowledgeable technician or facility manager, PUE is a useful indicator for points of further investment. However, there has been a recent trend of companies publicising their own PUE readings, sometimes under the banner of boosting their environmental credentials.

But publicising PUE can be problematic. Go to Google, for instance, and under its 'Going Green at Google' section it describes the company's 'Data Centre Efficiency Measurements'. Go to the page and at the bottom you will find Google listing its own PUE, which is a tasty 1.1-3, compared to the current trend of the sector of 1.9.

And herein lies the problem. By presenting PUE within this section, some argue that Google may be presenting the metric in an overly significant way.

"PUE is not a good metric for comparing facilities. It is more of a 'personal best' type of metric that can be used as a means of demonstrating improved energy efficiency," said Bill Kosik, green datacentre technology principal for HP's Critical Facilities Services.

But, by publicising the metric along with a detailed graph, Google appears to encourage transparency in PUE disclosure.

Kosik believes that PUE "cannot and should not be used to compare facilities " because it is "heavily dependent on climate, reliability level and many other factors".

So when Google lists its PUE as between 1.1 and 1.3, the firm is not displaying the range of factors; looking at the numbers doesn't tell you whether the datacentres are in a hot or cold environment, or whether their power comes from renewable or non-renewable energy.

This encourages you to think that, if a PUE is low, a company is 'going green'. This is a dangerous assumption, because it's a 'personal best' measure, not an ironclad figure that reflects all the input factors. For instance, if your datacentre is in a hot country, your PUE will be artificially higher owing to the extra electricity required to cool the IT infrastructure. But if your datacentre is in a cool area, such as those run by Microsoft in Ireland and Google in Belgium, your PUE will generally be lower, thanks to a cooler environment.

"[PUE] was designed to be a tool for improving infrastructure efficiency, and not as a tool to say that one site was more environmentally efficient than the next," said Bob Seese, chief datacentre architect at Advanced Data Centres.

James Cole, director of Keysource, a datacentre design and management company, explained that his company adopted PUE relatively early on, and that people had been "champing at the bit" for access to a simple metric.

But PUE falls on its own sword because its inherent simplicity - a single number - is what makes it easy to exploit. If I power my datacentre using a typical non-renewable energy source, in many cases my electricity bill will be lower and this will artificially lower the PUE rating.

Naturally, PUE can be pushed lower thanks to innovations in the sector, such as Facebook's recently announced bespoke datacentre in the US, reportedly costing $188m (£115m), or Google's bespoke shipping crate datacentres, but, again, these are site-specific innovations.

Kosik sums it up. "The electricity generation process is really the most important from a greenhouse gas emission perspective," he said. "It is fairly meaningless in the big picture if we design an extremely efficient datacentre that gets its power from an inefficient fossil fuel-based generation plant."

Cole added: "The analogy I often use is miles per gallon in a car. It's very similar. You get a very general indication of the effective use of petrol in a car but, if you look at a tractor in industry against an ambulance or a truck or a sports car or a hybrid car, you can't necessarily compare those miles per gallon because you undermine the effective use of that vehicle."

PUE's fundamental problem is that a lack of alternatives means that it becomes "very perverse in its desire to be more than it can ever be", Cole concluded.