How should software vendors calculate licence fees, and should charges be tied to the number of processor cores? With news that Intel will launch four-core Clovertown chips by the end of this year, such questions are increasingly pressing.

Companies should already be planning the transition to multicore processors – deciding how many to buy, when to schedule their arrival, and which applications or services most require the extra performance. And they must decide whether some applications should be left on single-core systems. Their choices will help firms to anticipate the affect of four-core processors on the cost of their software licences.

However, the issue of licence costs may be further confused if applications or services are automatically managed so they run on a variable number of processors as utilisation fluctuates.

The stance taken by some software vendors, including Microsoft and BEA, is to charge per socket regardless of the number of cores. Others, such as Oracle, calculate charges using fractional multipliers based on the types of chip used, which also causes complications for IT managers.

Meanwhile, IBM now bases its charges on Processor Value Units (PVUs), which may also make firms’ calculations harder. PVUs require the benchmarking of chips of all major vendors to come up with a PVU rating for each of them. Nevertheless, the PVU rating system does have some merit.

Many years ago, Digital advocated the measure of millions of instructions per second (Mips) as a way to rate performance. Ironically, 1Mips was only ever 780,000 instructions, but that is another story. Mips figures soon became the standard rating system for all computer makers.

The Mips ratings may have been approximate, but they gave IT managers a reasonable idea of relative performance.

Today’s licensing issues and complexities suggest that IBM’s PVUs could form the basis of a similarly valuable rating system to determine the cost of software licences. This approach would be especially suited to service-based, per-use business models – for example, the licence fee might equal the PVU rating multiplied by a cost per time unit, multiplied by the time used.

For widespread acceptance, this approach would probably need to attract the support of an industry consortium.

Meanwhile, as a manufacturer of processors, servers, operating systems and application software, IBM seems well positioned to promote this idea. And if IBM does the job well, it might lead to a business model on which service-based infrastructures are commonly built in future.