Dual-core Processors Challenge Licensing Models
ffub writes "Changes in hardware (such as dual-core processors and virtualisation) are making software licensing increasingly difficult for software firms. Companies still prefer the per-seat one-off license, while subscription models are favoured with software firms. But neither model reflects well the way software is used these days. The Economist looks at the situation and briefly touches on how Open Source could benefit from the muddle."
Amazing that this is turned into a problem. I've dual processor Macs for years.
Dual cores does not equal dual processors. If you had dual processors either your software had a dual+ licence, or if it knew it was restricted to a single processor, it would stick to that.
The tradeoff in terms of hardware and licence cost between dual CPUs and (faster) single CPUs was simple (infact there was little difference per-chip performance wise between them). There was also a good level of seperation between parallel and serial performance: with a reasonably wide physical space between the dual chips, tasks that could take advantage of dual cores were those with the most parallel nature, tasks that were highly serial couldn't get much uplift in performance.
Dual/multi cores is different in the sense per-CPU performance is faster for multi-core CPUs than single core CPUs, no matter what sort of task it is - CPUs with a single core simple can't beat multi-cores even on the most non-parallel type task. Dual core doesn't mean having 2x the throughput of a single core processor of the (otherwise) same spec, however: as some serials of instructions cannot be reduced to parallel, their performance will be 1-2 times a single CPU, probably in the upper quarter of this. As more cores are added the potential for parallelism reduces, and multi-cores increasingly underperform the sum of cingle core performance.
Now, this poses an interesting dilemma. Previously companies could add ever faster CPUs and (probably) get their software running faster and faster at no additional software cost. With dual+ core CPUs they're instantly doubling+ their licencing costs, while not getting a linear uplift in performance, and having previously expected this performance at no software cost. That's why they're getting upset.
The end result? My crystal ball suggests there will be a fixed cost aspect to software, perhaps as a scalar 0-1 times the number of cores on a CPU, accompanies by a per-CPU cycle charge over a hurdle rate.
I could talk for hours about parallelism vs. serialism, I hope the above breakdown wasn't too simplictic.