It's not often that such a thought crosses anybody's mind, but you have to feel sorry for enterprise software vendors. Exactly how are they supposed to price their wares?

Once, it was simple. Back when there were only minicomputers and mainframes, or even client-server, software tended to be used only by a firm's staff. It was therefore entirely reasonable to charge by the user.

Then, of course, came the web.

Counting users suddenly looked like a bad idea. Nobody wanted to have to buy a million extra licences just in case they had a busy month. So charging by the processor became more common. Of course this was problematic because unlike people, not all chips are created equal. Charging the same for Digital's Alpha as for Intel's Pentium Pro just didn't seem fair.

So some enterprising software vendors - Oracle, mainly - cooked up schemes to calculate pricing by an abstract universal processing unit. Buyers rebelled and eventually even Oracle was forced to concede that a processor is a processor. Which at least had the virtue of giving customers of expensive Risc systems relatively cheap software.

Today, a chip is starting to look a lot less like just a processor. Technologies such as Intel's SpeedStep, Transmeta's Code Morphing and AMD's 64/32bit dual-mode cleverness can give a single chip multiple personalities. Hyper-Threading can make one chip do almost twice the work. Multiple cores similarly ramp up the output. Combine several of the above and what do you get - a single processor or half a dozen?

As a result, AMD has already begun lobbying vendors to price by the socket - the physical connection between the processor and the motherboard - to try to make things clear-cut again.

This would be OK, were it not for current developments in software.

Software virtualisation allows one software system to host a variety of other software systems. So, for a hypothetical example, a dual-processor server running Linux could host three instances of Windows Server 2003, each of which could in turn host four instances of Windows 2000, with just one running an instance of software X. Does vendor X therefore demand a two-socket licence, or does the buyer assert that software X runs on one-sixth of a processor?

No doubt the average vendor will adamantly argue that the more expensive interpretation is the correct one, but buyers will only tolerate such attitudes while they remain reasonable. As the capabilities of ostensibly single processors continue to grow, and virtualisation allows for ever more flexible systems, buyers will demand prices that reflect the real proportional value of the software, not an inflexible view of potential capability.

So, where is all this heading?

It seems likely that many enterprise software vendors will eventually rip up their price lists and offer flat-fee, all-you-can-eat licensing. But even that is going to be complicated because most will want to charge large firms more than they charge small firms. And how do you measure the size of a firm? By headcount? Profits? Data throughput? Server processor sockets?

As we stumble toward this future, just a couple of things remain clear. One: IT managers with good negotiating skills stand to seize some great bargains in the coming months. And two: life could be a lot less complicated for those that use open source software.