Data center energy consumption is a huge issue. But it’s not really the big webscale players — the Facebooks and Googles of the world — that are at fault. They do a pretty good job building efficiency into their process and their gear. In aggregate, these “hyper-scale” data center operators account for less than 5 percent of total data center energy consumption, according to Pierre Delforge, director of the high-tech sector energy efficiency group at the Natural Resources Defense Council, an environmental advocacy group.
A couple of years ago, Google, for example, published a paper proposing that servers should be more like people; in that they go into super-hero action mode when there’s a big job to do and shut themselves down when there’s not. Of course Google (and Facebook and Microsoft etc.) have the resources to make that happen for themselves. It’s the smaller fry that are at issue here.
Idling servers at small data centers use too much power
To put the waste into perspective; data centers in total consumed 91 billion kilowatt-hours of electricity last year. That’s enough to power all New York City households for two years, according to the NRDC. So cutting waste — especially in the small and mid-sized shops — could have a huge upside. Paying for energy to keep servers idling is probably not the best use of anyone’s resources.
If nothing is done, according to a Delforge blog post: “data center electricity consumption is projected to increase to roughly 140 billion kilowatt-hours annually by 2020, requiring the equivalent annual output of 17 new power plants, costing American businesses $13 billion annually in electricity bills and emitting nearly 150 million metric tons of carbon pollution annually.”
The NRDC has a couple of recommendations that it says will help cut energy use.
- First, it wants to see a standard metric that would show average server CPU utilization. This would be a “simple, affordable, and adequate way of gauging data center efficiency that could be used immediately to drive greater IT energy savings in data centers,” according to the NRDC.
- Second, there should be incentives for data center operators to be efficient — and for customers to use efficient data centers. “Multi-tenant data center stakeholders should develop a “green lease” contract template to make it easier for all customers to establish contracts that incentivize rather than stand in the way of energy savings,” according to the report.
- And third, there should be more public disclosure of data center energy and carbon performance going forward. By everyone. Amazon Web Services, for example, is everyone’s idea of a webscale data center power, but according to critics including Greenpeace, it isn’t transparent about its energy efficiency or energy sourcing.
Assessing energy costs of the piece parts, as well as the whole
For context, many data centers use Power Usage Effectiveness, or PUE scores, to show the overall energy efficiency of their facilities. That is fine and dandy, but not enough, according to the NRDC.
“PUE rates the efficiency of the facility, not of the IT within it,” Delforge said via email. “There are many very efficient data centers which only have say 10 percent overhead for cooling and power distribution, but have their servers running idle or grossly underutilized. It’s like an airline flying a highly efficient plane empty or with less than 10 percent of seats filled: the carbon footprint per passenger would be very high.”
What NRDC wants is a simple metric to measure how efficiently the IT gear in the data center is operated. “This is complementary to PUE. And to run an efficient data center, both metrics are important,” he added.
Gigaom Research analyst Dave Ohara said the report brings up valid points, but more factors need to be considered. CPU utilization is just one metric, Ohara said via email. “RAM and hard-disks also use up energy … and can be just as underutilized … The problem is that IT asset management is mostly done as a bookkeeping exercise, not as part of a technical IT operations team who purchases, owns and operates the servers.”
Still, a single, uniform way to measure server CPU usage is a good first step down the road to better energy efficiency and transparency around that efficiency.