Enterprise software licensing is a mess. It has become fragmented, occasionally exploitative and certainly no longer fit for purpose. Take per-processor tariffs. For a while, when big iron ruled the datacentre, these sufficed as rough and ready reckoners of software value. Even the fact that vendors including IBM, Oracle and Veritas had been charging per processor core on IBM, Sun and HP CPUs went largely unnoticed because software was not assuming a disproportionate amount of the IT budget. It was only when Intel and AMD started to talk about plans for dual-core - and eventually multi-core - chips that alarm bells started ringing.

Per-core charging is an unctuous example of opportunist profiteering. A datacentre built on low-cost multi-core, multi-processor servers could potentially lead to huge hardware savings but rocketing software bills. Within a few years it will be almost impossible to buy servers without multi-core technology, so buyers would be penalised even if they did not want the extra mips these highly integrated chip packages provide. Dual-core chips do not even offer anything like double the processing performance, so buyers are hammered either way.

Oracle has at least put its head above the parapet to explain its thinking. Pricing and licensing vice-president Jacqueline Woods takes a bullet for her employer on the firm's web site and digs deep to proffer this unconvincing, bizarre explanation:

"Let me use an analogy. If you go to a restaurant and order two apples, it doesn't matter how the server delivers the apples to you. The apples could come on one plate or two plates. Either way, you will consume two apples. Processor licensing works the same way. Customers pay by the number of processors they use, whether they are delivered on one chip or two. Simply put, if customers require more processing power, regardless of its form, they also will need to increase their software licences."(

It may be an analogy but it's specious. Processing power today is achieved in many ways. Oracle should know this because it once charged by megahertz. Today's multi-core, cache-rich, HyperThreaded designs and new microcode instructions provide other ways to accelerate software performance.

Thankfully, competition will force Oracle and the other hold-outs to shift. Microsoft, HP and most of the open-source gang do not charge per-core so the tariffs for 10g, DB2 and other per-core products are going to look pretty silly stacked up against SQL Server, MySQL and the per-socket bunch. The mighty dollar will surely force a swift rethink in the coming months.

But there might not be a happy ending. Although Oracle has said it plans to introduce per-employee pricing in a move that would bring it into line with Sun, this won't be good for all buyers. Per-employee is in most ways preferable to per-CPU/core, but it will not be a great option for certain manufacturers, retailers, state-run businesses and others with high headcounts and thin profit margins. Per-named user (an Oracle alternative) and other models all have their pros and cons.

What is needed is more commonality between software licences to avoid gotchas, mind-numbing cost assessments or time-wasting bargaining sessions. But who is going to start the ball rolling?