Datacentre cooling and power consumption costs, boosted by higher-than-expected rates of server rollout, are leading datacentres into an economic crisis, according to a survey of 311 enterprise datacentre managers by the Uptime Institute.
Forty two percent of the respondents said their datacentres would exceed power capacity within 12 to 24 months unless they carried out expansion. Another 23 percent said it would take 24 to 60 months to run out of power capacity.
The managers reported similar figures for cooling: 39 percent said they would exceed cooling capacity in 12 to 24 months, and 21 percent said it would take 24 to 60 months.
The figures indicate that a sea-change in the economics of IT has occurred over the past few months, according to Uptime Institute executive director Kenneth Brill.
"A very significant change has occurred in 2006 and 2007," Brill said in a video presentation on the figures.
The figures indicate that datacentres are quickly becoming dramatically more expensive to run. For instance, to maintain cooling and power capacity, enterprises will be forced to expand, taking money away from IT budgets, according to Brill.
The shift began to occur in 2006, according to Brill, when power consumption rates for all but the bottom third began to grow significantly faster than before. While the bottom third's consumption dropped, the middle third double power usage rates to 10 percent and the top third more than tripled to 22 percent annual growth. The figures are based on a compound annual growth rate (CAGR) for 2006 to 2007.
This is due to server rollouts at significantly higher levels than expected and the fact that processor performance improvements are outpacing chip efficiency improvements.
In the US, the growth in datacentre energy consumption looks likely to outpace the rate at which power plants will be approved and built, raising the possibility that energy distribution to enterprises could be restricted, Brill said.
On an individual level, the power and cooling costs for a $3,000 (£1,491) server currently are set to exceed the cost of the server in less than two years, according to Uptime. Less than eight years ago, the value of a server was equal to 10 to 15 years' site costs.
Those eight-year-old servers are part of the problem. Uptime's figures show that enterprises have up to 30 percent obsolete, decommissioned servers in their datacentres that are no longer in use but are still using power.
Datacentre managers simply haven't tasked anyone with switching the servers off and disposing of them, something that should be part of any virtualisation project, Brill said.
Brill also recommended classifying applications into levels based on their importance to the business, and assigning each application to the lowest appropriate datacentre tier level.