eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
A new research study on data center capacity and energy efficiency has found that ever-increasing power and cooling costs are affecting the economics of IT more profoundly than many industry observers had thought.
The study, conducted by the Uptime Institute in Santa Fe, N.M., represented the views of 311 enterprise data center managers. The data was recorded over the last two weeks.
Forty-two percent of respondents indicated that their data centers would run out of power capacity within 12 to 24 months if they did no expansion. Twenty-three percent claimed they would run out of power capacity in 24 to 60 months.
Similarly, 39 percent claimed that their data centers would run out of cooling capacity in 12 to 24 months, and 21 percent claimed they would run out of cooling capacity in 24 to 60 months.
“Unless they have unlimited access to money, senior IT executives can no longer ignore [data center] power and cooling costs,” Kenneth G. Brill, executive director of the Uptime Institute and director of the research study, told eWEEK. Brill recently keynoted eWEEK’s Data Center Summit in New York City.
“We found confirmation for what we call the ‘Economic Breakdown of Moore’s Law,'” Brill said. “Basically, the increasing cost for power and cooling are fundamentally and almost invisibly changing the economics of IT.”
Virtually every respondent to the survey reported they were running out of power and cooling capacity, Brill said, which has resulted in a boom in data center construction and reconstruction.
“All money going into data center construction is coming out of IT’s budget,” Brill added.
Every 1U [smallest size] server incurs an annual facility cost of about $1,600 per year, Brill said. Of this, about $700 is for electricity. Since the street price for a volume server is in the $3,000 range, the electricity usage itself over four years will exceed the acquisition cost of the server, Brill said.
“This will occur even faster in high utility rate regions like New York City and the northeast,” Brill said. “On a total cost of ownership basis, power and cooling costs will exceed the cost of a server in less than two years. This is a radical change from less than eight years ago, when it took 10 to 15 years for site costs to equal server acquisition costs.”
Effects on the Enterprise Market
How do these trends affect the enterprise market?
“Something dramatic happened in 2006 and 2007,” Brill said. “Our members-who are all enterprise sites-had been installing servers at well under-predicted rates. In 2006, the past pattern was broken for all but the bottom third (in terms of power-consumption growth rate).
“While the bottom third actually reduced their power consumption, the middle third almost doubled their rate if power consumption increased to 10 percent annually; the top third more than tripled to 22 percent annually,” Brill said. “These are dramatic, budget-busting changes!”
The survey didn’t reveal reasons for the abrupt change, Brill said. “Virtualization was supposed to reduce the server growth rate, but this doesn’t seem to be happening yet,” he said.
The single biggest problem that can be addressed relatively quickly, Brill said, is in older servers that simply aren’t being used anymore.
“Unless they have been actively engaged in a de-commissioning program, most data centers have up to 30 percent comatose servers,” Brill said. “These are servers that have been replaced by new technology, but the obsolete hardware is still running because no one was tasked with turning it off and removing it.
“At $700 per year just for electricity, this is an expensive waste of both money and power/cooling capacity. Step one should be making an inventory of hardware that is suspect for being no longer needed and turning it off. All virtualization projects should have an explicit step of removing the old hardware at the completion of the virtualization project.”
What else can be done immediately to help save power and cooling costs?
“Applications need to be sorted into mission critical, business critical, business operational and business administrative reliability categories,” Brill said. “Each application then needs to be assigned to the lowest appropriate data center tier level.”