Data centre energy is a problem – and attacking Google, Facebook and Amazon won’t solve it, says Peter Judge
Last week, a US outfit called the National Resource Defense Council came out with a report on the energy used by data centres in the US, that’s got plenty to say – even to people outside America.
The report, Data Center Efficiency Assessment, is very well written, it has attention-grabbing infographics that put data centres in context, it drills down into the sector, and it makes suggestions about what should be done.
In a nutshell, it makes it clear that there is a problem – and that this problem is darned tricky. .
The giants aren’t the problem
NRDC sounds like a government agency, but it’s an environmental advocacy group. It’s got an axe to grind, but it is more sober and more business-friendly than Greenpeace, whose intervention in the IT world has ruffled feathers in the IT world in the past. NRDC commissioned the report from Anthesis, an experienced consultancy which believes that “commercial success and sustainability go hand in hand”.
The report starts with the problem statement: data centres use lots of electricity. If you add up those in the US, they consume twice as much electrical energy as New York, the largest city in the country.
That’s a good factoid, but not news. It follows up with a very striking pie-chart, which breaks down energy use between different sized data centres, and this may change your view of the cloud giants.
I’ve always known that cloud data centres are a minority of the world’s computing. They support the boom in consumer tech services, and they have the potential to subsume a lot of the stuff currently done by corporate IT, but it’s still early days. While a lot of stuff has moved to the cloud, a lot is still hugged close to the corporate bosom on internal servers.
I knew the process had not moved far, but the chart shows just how little the cloud has actually achieved in terms of dismantling corporate IT. Only four percent of the power used in US data centres goes to hyper-scale cloud services such as Google, Facebook, and Amazon. They’ve scarcely made a dent.
The co-location market (multi-tenant data centres) uses about 20 percent of the energy taken by US data centres, while in-house enterprise IT uses about a quarter (27 percent). The huge bulk, nearly half, goes to small and medium-sized data centres.
On TechWeek, we have written lots about energy efficiency at Google and Apple and Facebook, but it seems this is only the tip of the iceberg of energy usage. And while NRDC is talking about US, I believe the same holds true elsewhere.
Even if those big-brand cloud data centres weren’t already operating at close to optimum PUE (power usage effectiveness), and could slice their power use by half, that would still only take a sliver out of the overall power use of data centres.
So my first conclusion is that, if we want to make efforts to reduce data centre energy use, we have to start with the small people – corporations with small data centres of their own.
How do we control small data centres?
My second conclusion is that we don’t know how to do this. We can apply legislation, but look at the best example I know – the UK’s Climate Change Agreement. It only addresses large co-location sites, leaving out enterprise IT and smaller sites completely. Also, it doesn’t go far enough in cutting energy use, because it takes PUE as a measure of efficiency.
NRDC says the real waste of energy is idle or “comatose” servers, powered up and doing nothing, or under-utilised. A good PUE score tells us nothing about this source of waste, because it only measures how efficiently a data centre delivers energy to its IT kit, not how well the processing power is used. A completely idle data centre can still have a great PUE.
The report has a lot of great discussion about why servers are idle, including over-provisioning and the failure to implement power management, and extending to the economic model.
But how to change this? NRDC’s recommendations call for an efficiency programme based on a new server utilisation metric. This would be great, but there’s been a lot of discussion within the data centre community on the subject already, and it will be really, really hard to get that right.
The Green Grid’s DCeP, to take just one of the many approaches to the problem, admits you can’t judge how much useful work is being done by a server (because it depends on your definition of “useful”), and hands the problem over to the industry sectors concerned, to make their own standards up.
That’s honest, but it’s never going to provide a basis for the kind of solid program of incentives or regulations that NRDC wants to see.
Like I say, this report has made two things clear. Server energy is a big problem. And it’s a difficult one.
A version of this article appeared on Green Data Center News