A New York Times Sept. 22 article now causing heated conversation in the cloud computing community paints a gloomy picture of the average data center as a wastrel and energy glutton. In the eyes of the writer of the initial piece "The Cloud Factories: Power, Pollution and the Internet," data centers are prime suspects in the ongoing process of environmental degradation.
I am in total accord with the intent of the Times piece, seeking to make data centers more efficient. But I am dismayed at the casualties that the writer, James Glanz, inflicts as he goes about making his point. Surely, he realizes that the first part of the story describes an aging, mixed-system data center straight out of the 1980s or 1990s. Then he goes on to say that both the enterprise IT manager and cloud data center creator are culprits (two very different things lumped together). And, he concludes that the Internet user is an eager, witless accomplice to them both.
Either the Times underestimates the intelligence of its readers or it has lost touch with them. After all, it delights in publishing stories about the iPhone, and there are few drivers of data center construction that are equal to the use of smartphone apps. I don't remember reading about the dark side of iPhone 5 in all those iPhone 5 advance articles.
It's difficult, but what we should be trying to calculate is how to keep the economy growing with digital services, while protecting the environment.
Everyone is doing a lot more computing, as the story notes. But as we do so, the amount of electricity consumed per unit of computing is going down, which the story somehow misses. Nowhere does the Times address this salient point. Instead, it concludes we are doing a lot more computing and, therefore, we are all guilty of driving environmental degradation. If you're going to reform the world, you need to build a better soapbox than this.
[ Want to learn more about how Columbia Sportswear's cloud computing plans? See How Columbia Sportswear Will Survive Next Tsunami. ]
There's a lively discussion going on regarding these points at the New York Times' collection of related opinion columns, Big Data's Big Costs. "Is the convenience of cloud computing worth the energy and pollution it requires? Or should we wean off of 24/7 access?" are some of the questions asked.
Let me start with some contrary evidence. InformationWeek reported a year ago that the Environmental Protection Agency had predicted that between 2005 and 2010, the electricity consumed by data centers around the world would double, based on the rate of new construction. In 2011, the EPA had to back off that prediction because during that period, data center electricity consumption had increased by 36%, not 100%. The author of the study for the EPA speculated that the pace of data center construction had slowed, due to economic recession. I for one have driven around Silicon Valley looking for staked-out, but not-yet-built data centers; they're more scarce than Samsung handsets on Apple's Infinite Loop.
While some data centers were mothballed during the recession, most were built, and were much more efficient in using electricity than their predecessors. For example, PUE is the ratio of energy imported by a facility to the amount of power consumed by IT computing devices; it stands for Power Usage Effectiveness. The 1990s data center, with air conditioning pouring through a raised floor, and featuring a water-cooled mainframe at the center, had a PUE of 1.92 to 2. That means it consumed about twice as much power as the amount used in actual computing; cooling was the largest power consumer after computing itself.
Today, modern data centers built by Microsoft, Google, Facebook, or Yahoo have reduced that ratio to somewhere between 1.22 and 1.07--the latter mark set by Facebook's new Prineville, Ore., facility. This type of modern data center does so by delaying the stepping down of power line voltage until it's close to where servers and switches are running; that reduces the loss to resistance in the line. Instead of air conditioning, the modern data center uses ambient, outside air cooled by evaporation; the air is piped to the cool side of servers, blown into channels between baffles that steer it to the hottest components. Then it's collected in a hot aisle and flushed from the building. The process uses a fraction of the electricity of air conditioning.
The Times article doesn't mention PUE. It does cite a McKinsey study, commissioned by the newspaper, that found across several industries, the average server used only 6% to 12% of its available CPU cycles. To me, this describes the data center running one application per server, lest one application's activity overwrite the data of another (which causes a crash). In the past, InformationWeek has cited such a practice as utilizing 7% to 15% of the server, with the rest of the CPU cycles idled away. The practice reflects an IT manager's fear of downtime. At one time it was a realistic calculation that the business could afford the extra electricity, but it couldn't afford to go off the air for a few minutes or few hours.
But to some extent, the Times is busy pillorying a straw man. I don't know how to explain the McKinsey figures at this late date, but it's now 2012 and the times are a changin'. Maybe with his focus on the big picture, the writer missed the trend toward virtualization within the enterprise data center, where each server is loaded with 15 to 20 applications, or more. They run under lean hypervisor software, which manages CPU, memory, and I/O limits to avoid application conflicts. At sites where virtualization is used effectively, half or two-thirds of the number of physical servers formerly in use are now gone.