Datacenters are no longer glass-house sanctuaries whose capabilities can be measured by their floor space, and datacenter managers can no longer assume they are the masters of a corporation’s information technology simply because they understand how it works, according to a report released Oct. 23 by Gartner.
Datacenter managers have to stop thinking and planning for their datacenters either in the ways they were originally taught, or the ways they later discovered might work. Instead they have to plan for the infinite datacenter, which is defined by the business-process functions it delivers and power it can provide, not by the amount of floor space, legacy systems, headcount and other traditional ways of measuring the size and importance of departments within a corporation, according to David Cappuccio, research vice president at Gartner, in a public statement.
Creating budgets for the future using capacity planning methods based on space “is based on two flawed assumptions,” Cappuccio said. “That the existing floor space is already being used properly and usable space is purely horizontal.”
In a poll of 1,000 IT managers posted in August, the Uptime Institute reported that 67 percent of respondents reported fewer than 5 percent of servers in their datacenters were either unused or were targeted for shutdown. Forty-three percent don’t do scheduled audits to check the status of those servers, either for current use or planning capacity for the future.
It’s possible floor space is the only metric many datacenter managers have to work with, however imprecise it may be.
Genuine estimates of capacity – physical capacity, leaving aside virtual machines and infrastructures for the moment – should be measured in capacity per square foot or kilowatt, both of which take better account of high-density server racks with hundreds or thousands of physical servers.
Datacenters that grow at 15 percent per year would need to quadruple their floor space in 10 years, making cooling, power distribution and almost everything else less efficient rather than more. It’s more efficient to assume the same increase in capacity without assuming any increase in required floor space, which can be done by thinning servers down to 1U rather than 2U, or, in terms of blades, increasing rack capacity and other ways to increase the density, rather than the size, of a compute farm.
“An initial investment in planning time and technology refresh can pay huge dividends in the mid-to-long term for businesses anticipating a continuous growth in computing capacity needs,” Cappuccio said. Cloud computing and virtualization won’t prevent the need to plan actual increases in power, but can make the pace of growth more manageable and make it less likely that unexpected spikes in demand will be disasters rather than annoyances.
Image: Shutterstock.com/ Dan Collier