Which should come first, the grid or the app virtualization?

While DataSynapse FabricServer integrates with its sister product, GridServer, and the two are used to good effect together at Wachovia Bank, a grid is not required to use FabricServer, says Jamie Bernardin, DataSynapse CTO. Tony Bishop, senior vice president and director of product management for Wachovia, in Charlotte, N.C., agrees.

Bishop advises network executives who have not yet embarked on any utility-computing infrastructure to start with the application server virtualization rather than a grid (i.e., CPU virtualization), even though the grid came first at Wachovia. Grids tend to require a lot of custom coding when moving applications to a multi-CPU environment, he says.

In contrast, the FabricServer environment is easily dropped on top of an existing application and is "not offloading processing of services. All it is doing is adding dynamic management to the actual J2EE environment," he adds.

And while your team gains experience in virtualizing another layer in the infrastructure stack, it could be racking up some serious costs savings by reducing server sprawl. This is true even for organizations that have consolidated servers by virtualizing the operating system layer using products such as VMware ESX Server. Whereas server consolidation at the operating system level might eliminate 10% to 20% of a company's servers, consolidating application servers could eliminate as much as 50% to 60%, Bishop contends.

VMware creates a first tier of server virtualization in which the supply of servers is divided to get more use out of each server. Application server virtualization then increases server efficiency by keeping more of your partitioned servers busy and less of them idle. As Bishop says, "Both strategies are important."