To automate tasks such as patch management, server provisioning and installation, you have to centralize operations, says Delaware North's Armstrong.

Using one server per application, Delaware North built a data center in 1999 that it expected to last for 10 years. But the center was swamped by August 2003. Processing a request for a room at its hotel properties in Yosemite National Park took 15 to 29 seconds via the company's Web site. Unbearable, to prospective customers.

"We didn't have the equipment, processing [power] or space" to expand further, Armstrong says.

The answer was virtualization, which made consolidation of data center operations possible.

Using software from VMware, the company could get by with one server where seven used to stand. Space needs were cut in half. Maintenance costs fell from $28,500 a year to $1,800. Capital investment dropped by $105,000 a year. Provisioning a new server takes one hour, instead of eight. And processing a reservation takes seven to nine seconds.

Armstrong says the approach can be replicated for storage and network systems. He is adding three more servers—and expects to run as many as 25 applications on each.

Today, 25% of the company's data- center systems run on shared servers; that figure should reach 50% in the next 18 months. Where Delaware North has put one to seven people to support information technology at as many as eight properties, the goal is to have the 54 people at headquarters in Buffalo, N.Y., handle all the support.

Next up: Mapping out data center processes that can ultimately be automated. Inflow's Monsour says documenting each step in help-desk queries, test server requests and patch installations can take anywhere from six to 18 months. And those are the easy ones. "You have to start small with something like patch management and then work up to figuring out how to automate something like security policies," he says.

So, it pays to get started early.