Tracking data center energy costs
Ever since the data center became the poster child for IT energy waste, companies have been looking for ways to make it more efficient. Nobody knows that more than Alan McNab, president of consulting firm McNab and Associates, who has been working with a large financial services firm in New York on a project to identify and monitor cost drivers in its data centers.
Operating expenditures over the lifetime of data centers can be two times the amount spent to build them, and yet most companies have very little concrete information on what's driving those operating costs, says McNab. "There's a real need to get more cost transparency around operational expenditures," he says.
If a company can understand what's driving operational costs, then it can make more intelligent decisions about how, when and where to deploy some of the new cost-saving technologies like virtualization, he notes. In addition, the company can use cost data to design more cost-efficient centers in the future.
As a first step toward identifying data center operating costs for the New York financial services firm, which he declines to identify, McNab is running a trial program using wireless sensor networks from Arch Rock to identify temperature and other metrics in specific areas of the data center. At the same time, McNab is working to tie in all sorts of other data that is relevant to operational costs, including the model and age of specific servers, when each was deployed, how it's being used, how much electricity it uses, and how many trouble tickets have been issued for that server.
Arch Rock has specifically designed its wireless sensor products for integration into IP networks via Web services. "The sensor is an IP device on the enterprise network," explains Arch Rock CEO Roland Acra. "IT managers can access the sensors over the Web and manage them just as they would any other type of IP device. The IT person can pick their programming language, their development environment, their operating system."
That means IT staff can program the devices to do what they want without having to learn a proprietary language or go through the operations or manufacturing specialist, he notes. "And they can make calls into the sensors through Web services directories just as they do with databases and other applications in the IT infrastructure," Acra says.
Ultimately, integration and analysis of all this information could enable data center operators to cut their costs substantially. If one area is running too hot, for example, a building-automation system could open or reconfigure vents to increase the air conditioning in that one area only -- rather than unnecessarily cooling the entire room. Or if the data center uses virtualization, certain applications could be moved from hot aisles of servers to cooler aisles. Right now, however, it's still a far cry from being integrated into the IT infrastructure.
"This stuff with the sensors, this is way out there," McNab stresses. "It's not happening today."
Even though companies have the ability to track power use down to the level of the individual power strip, few if any are collecting that data or integrating it into the enterprise yet, agrees Joshua Aaron, president of Business Technology Partners, a New York-based IT infrastructure consultancy. Some clients are beginning to ask about it, however.
"A lot of companies are collecting a lot of data, but not a lot of companies are sophisticated to the point of using it for real-time data mining or iterative planning. It's not usually integrated in real-time applications where people are looking at it in relation to other things every day," he says.
But Arch Rock's Acra says companies could do cost accounting down to the level of specific servers that are performing specific applications. "Through sensors, you can get electricity consumption data from a cubicle, a rack of servers or even each individual server," he says.
"But you don't want to take that data in isolation. You'd want to tie electricity function to particular business functions," Acra continues. "In a financial firm, for example, was this energy used by traders who were doing a lot of activity? Or was it the quantitative guys who were doing technical modeling on stocks? Or were the investment bankers banging away at M&A?"
If energy costs keep going up as fast as they have in the past year, such sensor applications could well be among the first to link up with enterprise IT. No one right now seems to know what sensors' killer app might be. What's clear is that the ubiquity of sensors is producing lots of data that companies would like to use in more ways.
"The data coming off of these sensor networks is becoming more important in overall business management, whether that's asset tracking, whether it's just-in-time manufacturing, or whether it's connecting electric demand to the supply side," says Cisco's Platon. "The last 20 years we've spent connecting some 4 billion computers. The next 20 years are going to be spent connecting some tens of billions of industrial objects."