The shape of modern computing is changing rapidly. It's easy to get carried away with the technical details — visions of massively multicore chips embedded in a sea of virtual computing power seem exotic, exciting and daunting. Yet the underlying motive force behind IT isn't the flow of electrons through a logic gate or the capturing of meaning from data. It's money.
If highly parallel, highly virtual computing demands new technical design and programming skills, it's just as testing for software vendors' business models. The old idea of one box, one piece of software, one licence has been stretched to breaking. Attempts to switch to per-core schemes have met with resistance from users who feel, quite rightly, that they are being arbitrarily penalised for performance increases that in the past came for free with clock-speed upgrades.
IBM is building a new scheme for the future. It has created "processor value units" (PVU), a rating scheme of staggering opacity that prices licences per core, but varies that according to the class of chip. Any single-core chip has a PVU of 100, as do individual cores for RISC or Itanium chips — single-core Itanium is 100, dual 200. Devices such as dual-core x86 devices are 50 PVU per core, and Sun's T1 octicore is 30 PVU — making that chip 240 PVU in total. Software is then licensed at so many dollars per PVU — multiply that by the hardware's rating, and you have your licence fee. That's the equation at the heart of IBM's proposition.
The interesting bit will come when IBM starts to assign PVUs for more abstract entities, a step on which the company is currently silent. We can guess: a virtual x86 might be rated at half the underlying hardware, for example. It will even be possible in theory for software to measure the PVUs of its environment and set its own costs. IBM has said that the end position will be a pay-per-use model, where no matter what you do with software you'll pay for the work you get out of it.
As it stands, that has a chance of being fair and flexible. It also has the chance of bogging down in complexity, creating a licensing environment which is difficult to understand, hard to manage and impossible to cost. It could penalise efficiency by over-pricing effective usage models, it could discourage hardware upgrades and make cross-company price comparisons impossible.
Alternatively, it could also encourage efficiency by profiling usage and encouraging good allocation of hardware resources, even create a competitive market for hosted virtualised services. Run your database on our system, and save on licensing.
For this scheme to be a success, IBM must be much clearer about how it works, how it sets PVUs, whether it plans to preferentially value its own hardware, and what the roadmap is for future developments. Users must be able to decide how it will affect their plans, and IBM must be receptive to their concerns and ideas. The overriding need for all parties is clarity: the last thing anyone needs is a collapse into rapacious confusion akin to mobile phone service pricing. We said that the underlying motive force for IT is money: that's not strictly true. The real driver is trust between vendor and user. That's the equation IBM ignores at its peril.