Data Centers (DC’s) – they are hungry beasts. Ten years ago the EPA estimated that DC’s consumed 61 billion kilowatt-hours of electricity in the US. Just three years ago the National Resources Defense Council (NRDC) was calling the consumption level at 91 billion kilowatt-hours – almost a 50% increase in six years. Let’s put this into perspective. That power consumption is enough to power all the households of New York City twice over. Think of it, in 2013 this represented about 2%+ of all the power produced in the United States.
The prognosis was even worse. The NRDC projected that data centers were going to consume 140 billion kilowatt-hours by 2020, another 50%+ increase. But happily a just published new report by the Lawrence Berkeley National Laboratory (Berkeley Lab) shows that data center power consumption has essentially plateaued and will grow only slowly despite a huge increase in the number and size of data centers and a 40% increase in the number of servers from 2010 to 2020.
Good news but what happened? The Berkeley Lab discovered that around 2010 DC energy consumption flattened out at 2% of the total US consumption. (Overall energy consumption in the US has also roughly plateaued since 2010.) How can that be when we know that data center building is going gangbusters?
The key is that data centers for cloud providers (called hyperscale or web scale) are much more efficient than the legacy enterprise data centers. The Facebook’s, Google’s and Amazon’s of the world run tight ships. As well they would, since power and cooling in an old-school enterprise data center can constitute up to 24% of the annual budget.
A number called the PUE or power usage effectiveness measures data center power efficiency. A PUE of 1.0 is considered very desirable while those of 2.0 or 3.0 are very inefficient. Google publishes its PUE at 1.12 and Facebook has said that one of its centers has a PUE of 1.07. The federal government would like to get to 1.5. Being clever about how you design, operate and cool the machines is how you achieve this kind of efficiency.
Definitely this is goodness and points to how adopting cloud supports a green policy and objective. Unfortunately, there are still a lot of legacy data centers. The Berkeley Lab study points to over 60% of data center power consumption coming from smaller facilities. A recent IDC report shows that more than two-thirds of the traditional enterprises it surveyed logged a PUE of more than 2.0, and 10 percent were over 3.0 or didn’t know.
Cloud services are increasingly being adopted and replacing the old on-premise paradigm. The initial effects of this are being seen in the positive move to energy conservation in IT and so far we have good projections of limited power consumption growth for the next several years. That’s certainly good news but consider how much better it might be if the shift to cloud was faster?