Tech public relations likes to talk about the cloud as though it exists in an entirely conceptual form – files, apps, and even computing power existing merely in potentia, flowing mystically to you from some digital Olympus. It’s a neat vision, and not, perhaps, an entirely unfair one, as far as it goes.
But the fact remains that all computing is done on silicon that exists somewhere in the real world, no matter how many layers of abstraction it goes through. If it’s a computing task, there is a real computer, somewhere, pushing 1s and 0s around to make it happen.
So where are those computers?
In the case of the private cloud, the answer is, of course, simple – they’re in your own data center. For the public cloud, however, the question is a lot more complex, and the answer is hazy.
For one thing, a lot of the actual hardware used by the big public cloud providers in the U.S. lives in collocated facilities alongside servers used by other companies, making it difficult to pin down exactly whose computers are whose without a detailed look inside the buildings.
That said, there are stand-alone facilities – massive data centers run by single providers – where, it seems safe to say, much of the public cloud’s heavy lifting is done. North America accounts for a huge proportion of the public cloud services consumed worldwide, according to IDC – the sector saw $32 billion in spending in 2013, and the research firm’s analysts predict a compound annual growth rate of 20% through 2018.
So, given that Microsoft, Google and Amazon are largely considered to be the big three of cloud services in the U.S., we thought a map of their major data centers would go some way to answering the question of where, exactly, the cloud lives.
We asked all three companies where their cloud hosting resided, and got varying levels of cooperation in response. Amazon flatly declined to provide any information beyond a list of edge locations and service zones. Microsoft offered a partial list of municipalities where its cloud data centers are located, but said that it didn’t want to provide a complete list “to ensure a high level of security for its customers’ data.” Google, on the other hand, publicly lists the general location of its data centers on the web, making identifying them via satellite imagery a relatively simple task.
Map of biggest public cloud vendors’ U.S.-based data centers.
(Map key: Red = Amazon, Blue = Google, Yellow = Microsoft. If you are having trouble seeing the interactive map, click here.)
We found 13 stand-alone data centers, located across the country, and there are probably several more out there, at least in Amazon’s stable. The most popular region seems to be the Southeast, with six facilities that cropped up are in Virginia, the Carolinas, and Georgia. The stretch of the Columbia River that forms part of the border between Oregon and Washington also has at least two data centers along its banks. Both Microsoft and Google have a presence in the state of Iowa.
Picking a location for these data centers is dependent on several different factors, but the most important is the cost of energy – the Columbia River, for example, is an attractive site in part because of the cheap, plentiful hydroelectric power generated there, which Gartner infrastructure research chief David Cappuccio said can range as low as $0.02 per kilowatt. In addition, land in the remote Columbia Valley is inexpensive, and there are tax incentives available.
It all revolves around minimizing costs, he said. A several hundred thousand square-foot data center can cost more than $1 billion in initial capital, and more than $80 million a year in operating expenses. That’s not including the costs of half a million gallons of water a day (though that’s dependent on an individual facility’s design) and $11 million a year in electric bills. (Also read "The world's coolest and greenest data centers".)
This is part of the reason that the companies that build the data centers are generally a bit secretive, he said.
“Operating efficiencies are a competitive advantage – saving single percentage points has noticeable impacts on monthly operating costs, which at these sizes can add up pretty quickly,” Cappuccio told Network World.
Those efficiencies do vary by region, he adds – the Southeast has low energy costs and a lot of existing competition, which means that there are experienced construction crews and a skilled workforce already available. Iowa is attractive because of low land costs and tax incentives.
The technique used to identify the data centers on our map centered on Google Earth imagery. Once we had city and state information, we combed the location for buildings that featured telltale infrastructure like generator fuel tanks and heavy-duty cooling systems. In the case of Google data centers, it was often possible to see signage plainly identifying the facilities as Google’s from street view, while Amazon and Microsoft required a comparison with public records.
This brings us to an important caveat: While our identification of Google’s data centers seems relatively clear-cut, we caution that our information on Amazon’s and Microsoft’s facilities is indicative but not entirely conclusive – neither company would confirm or deny the information when asked. So while the balance of probability seems to be that the listed facilities belong to Amazon and Microsoft, bear in mind that these are preliminary identifications, not definitive ones.
Amazon, in particular, likely has data center holdings far more extensive than the map might imply
This story, "Mapping the Cloud: Where Does the Public Cloud Actually Live?" was originally published by Network World.