Adams wants one machine to look like many machines, but it is just as possible to virtualize the other way: making many machines look like one. Virtualization underlies the well-known RAID storage tricks that allow many disks to be treated as one huge drive for ease of access, and one disk to be treated as many for the purpose of robust backup. Another prime use for virtualization is development. The hardware world is growing much more complex all the time: Product cycles are turning faster, the number of device types is always rising, and the practice of running programs over networks means that any given program might come in contact with a huge universe of hardware. Developers can’t begin to afford to buy all of this hardware for testing, and they don’t need to: Running products on virtualized models of the hardware allows for quality assurance without the capital expense. Virtualizing the underlying hardware also gives developers far more control. Peter Magnusson, CTO of Virtutech, a systems simulation company in San Jose, Calif., points out that you can stop simulated hardware anywhere you like, any time you want to investigate internal details.

Unreal Future

During the next year or two, virtualization is on track to move from its current success in storage, servers and development, to networks and data centers. So CIOs will then be able to build software versions of firewalls, switches, routers, load balancers, accelerators and caches, exactly as needed. Everything that was once embodied in cards, disks and physical equipment of any kind, will be organized around a single point of control. If virtualization vendor promises materialize, changes that once were out of the question, or that at least would have required considerable man-hours and operational risk, will be done in minutes, routinely.

What those changes will mean is very much a topic for current discussion. For instance, all the new knobs and buttons virtualization provides will raise issues of policy, because it will be possible to discriminate among classes of service that once had to be handled together. You will, for instance, be able to write a Web server that gives customers who spend above a certain limit much better service than those who spend only half as much. There will be huge opportunities for automation. Infrastructure may be able to reconfigure itself in response to changes in demand, spinning out new servers and routers as necessary, the way load balancing is done today. (Certainly IBM et al. have been promoting just such a vision of the on-demand computing future.)