The Art of IT: Virtual Reality

Remember when Dad told you there's no such thing as a free lunch? He was right--there isn't. It's particularly so with virtualization. Every time you hear the term applied to a new technology, your first question should be, What will this new virtual benefit cost me?

In a steady state, virtualization is almost always a bad thing. Virtual memory is the original form of virtualization as applied to computers, and it effectively revolutionized how computers do their jobs, but at a significant cost. Virtual memory lets every program view its memory as one contiguous space, with no gaps and no holes. The space can even stretch beyond the physical memory. Before virtual memory, program loaders had to find a section of memory that suited the app's needs. After running a few apps at once, memory became fragmented and the loader couldn't find available memory slots big enough for applications. Then it was reboot time--DOS, anyone?

Virtual memory is a good thing today because we take for granted that we'll be starting, stopping and moving programs all the time--the OS does it thousands of times per second (and hence no steady state condition ever exists). But we pay for virtual memory in performance. To virtualize memory, the OS builds a map that correlates physical memory to the virtual space the application sees. Every memory access requires a map lookup, which takes both time and hardware circuitry.

It's similar with storage and processor virtualization. If we never needed to move or grow storage volumes, storage virtualization would be an unnecessary encumbrance. But needs change and volumes must move or grow. The trick is to find the storage virtualization technology that gives you mobility at the lowest possible cost, in terms of both complexity and performance.Not Quite Ripe

You can tell that processor virtualization is only now becoming interesting because it hasn't fully addressed the need to move OSs and apps fluidly between systems. Virtualization technology by itself is necessary but not sufficient to develop new uses for pools of processors. For example, it wasn't the advent of memory maps that made memory virtualization useful; it was the development of operating systems that could actually make good use of virtual memory maps that caused a sea change in computing.

Two initial uses for processor virtualization have proved popular: Developers can quickly test new apps by using virtual production environments on their development machines; and IT architects can use it to consolidate lightly used servers. Both are beneficial and can tolerate the trade-offs demanded by processor virtualization--a substantial performance hit, particularly for I/O, and significantly greater memory and storage requirements.

But the real sea change will come about when we have management tools that can take advantage of the hardware abstraction and resource protection that comes with processor virtualization. The vendors now understand this, and are jockeying for position by nearly giving away the hypervisor software that does the hardware abstraction and memory and I/O management. This is being driven particularly by the hardware virtualization support that AMD and Intel are building into their chips.

Because the potential for processor virtualization is so great, examples of its ultimate use sometimes feel trivial. One example: a company that needs 12 exchange servers running at 9 a.m. when everyone gets to work, but needs 12 instances of an inventory analysis program running at 5 p.m. after all the day's sales are in. The company benefits if it can have just one set of hardware to meet both needs.The true value of processor virtualization technology is yet to be understood. For now, make sure you understand the costs, and take the benefits as they reveal themselves.

Art Wittmann is editor in chief of Network Computing. Write to him at [email protected].