In its fourth year as a public company, NuVasive, a supplier of devices for minimal-invasion spinal surgery, pulled in revenue of $250 million last year, up 62% from the year before.

In order to sustain this rapid growth while keeping expenses in hand, it turned to VMware to virtualize one of its most important applications, Microsoft Exchange Server. Bill Moore, IT infrastructure manager, opted to put his Microsoft Exchange Servers into four VMs after running it in the startup years on a single physical server.

In order to achieve built in failover and recovery, he split the four VMs across two dedicated Hewlett-Packard blade servers in his San Diego data center, each with four CPUs and 32 GB of memory. They're governed by VMware's High Availability product to manage failovers.

With the upcoming implementation of VMware Site Recovery Manager, he will get full off-site backup and recovery by adding two servers at his disaster-recovery site in Memphis. To get all that redundancy the old way, he would have had to buy four four-way servers for each location, he said.

"We've just implemented Site Recovery Manager. It goes live at the end of the month," said Moore, giving him the second level of redundancy that he's always wanted for his Exchange Server users.

Exchange isn't everybody's first choice as an application to virtualize. Exchange users depend on access to e-mail and tend to become vocal when they don't have instant access. In many shops, there's a tendency to regard Exchange as not only deserving its own physical server but also as a standalone, not-to-be-meddled-with object.

NuVasive changed all that when it migrated from Exchange Server 2003 to Exchange Server 2007 at the beginning of 2008. Moore now runs VMware High Availability as part of his Virtual Infrastructure 3 environment in San Diego, and with only two VMs per physical server, he has stayed within VMware's recommended best practice of using less than 50% of a High Availability server.

That guarantees constant machine capacity if one server has to take on the workload of both. He figures he can even add a third VM in the future to each blade without hitting the 50% mark.