According to conventional wisdom, if you already own a mainframe, it makes sense to keep it doing what it's doing. But for new projects, like setting up a high-power server farm in a cluster or grid configuration, go for something with a lower per-unit price tag.
There are exceptions. The Advanced Computing and Information Systems Laboratory (ACIS) at the University of Florida bought an entirely new mainframe to research virtual machines and the performance of Linux clusters. Why?
Admittedly not to save money. ACIS was only able to buy the mainframe because it got steep educational discounts from IBM, and, even so, bought a bottom-of-the-line z800 mainframe, according to Renato J. Figueiredo, an assistant professor of electrical and computer engineering who helps oversee the project. Even at a relatively high price, the mainframe was the best machine on which to study the viability of virtual machines because of its ability to support so many. Virtual-machine software is a way to make one server into many, by essentially dividing one server's capacity into many smaller chunks and running a separate operating system on each. Mainframes can host thousands.
It's that ability to essentially replace dozens of Linux servers that makes a mainframe cost-effective in new roles, says Mike Kahn, managing director of consultancy The Clipper Group. Grid computing is just a way to provide computing cycles, regardless of the operating systems or hardware on which they run, a function at which the mainframe excels. "It's old, ugly stuff, but if it hides behind a curtain and looks like another machine, you don't care if it's a mainframe or a giant Unix box," he says.
Projects: Management 
