Muticore chips challenge per-socket software licensing
Subscribe now for $100 (23 issues) and save more than 37% off the cover price!
Get the latest news from Computerworld delivered via email.
Sign up now
The arrival of eight and 12-core x86 chips may make per socket software licensing seem like a good deal for Microsoft server software customers. Why? Because the new multicore technology promises to let them increase chip capacity while theoretically reducing the number of sockets.
At this point, Microsoft doesn't seem particularly worried that its server licensing revenue will melt away as its customers move to the increasingly powerful chips. However, analysts wonder how long that will last as chip makers add more and more cores to each processor.
Indeed, Microsoft officials say there are no plans to change the per socket pricing approach the company uses, even as AMD promises to start shipping 16-core chips next year.
Eric Jewitt, director of Windows server marketing at Microsoft, says the company looks at core increases as a continuation of Moore's Law, a doubling of capacity every 18 months or so.
A company could increase compute capacity by migrating from quad-core to 12 cores, which would consequentially reduce its sockets and software licensing costs.
However, that only works if the user is taking a mirror of the database onto the new chips, says Jewitt. While chip capacity increases, so is the amount of data, roughly doubling every year and creating a need for more capacity, he says.
The gains in compute power aren't necessarily matched by software.
As chip makers add more cores per chip, it creates a need for software that can perform in increasingly parallel environments, says Scott Rosenberg, CEO of Miro Consulting, a Woodbridge, New Jersey-based firm that advises companies on software licensing issues. "The software industry needs to catch up right now to the hardware industry," he says.
Still, the chip vendors are making a point of showing that users can reduce the number of servers by moving to their high capacity chips.
Not all vendors use the per-socket pricing schedule.
Oracle, for instance, instead assigns a value for each by chip manufacturer. In the case of AMD and Intel chips, for example, Oracle has set a processor factor of .50 per core, meaning a quad core is counted as two chips.
Analysts wonder if per-socket, or even per-core pricing, will survive over the long term.
On the issue of the per-socket model, Rosenberg believes that as core counts increase it is inevitable that Microsoft will move to some other pricing scheme. Per-core pricing, he says, "has got to be a consideration."
Christopher Voce, an analyst at Forrester Research, says software vendors may start charging more for supporting tools to offset the impact of the increasing cores. He added that with the increasing abstraction of the hardware layer to virtualised environments, it will be increasingly harder for vendors to price based on physical assets.
"The traditional approaches to licensing won't hold up," says Voce, who sees the industry moving to business-based metrics, such as per-user pricing.
Northern Illinois University IT director Walter Czerniak says based on 40 years of experience in the business, he doesn't expect that vendors will lose money because of chip improvements.
"If any vendor starts to lose too much revenue based on hardware technology, it won't take long to change their software licensing agreements,"Czerniak says. What keeps them in check is competition, he added.
Scott Archibald, managing director of Bender Consulting, a Houston-based management consulting firm, says he expects companies running mega datacentres and large cloud providers to quickly adopt the new large chips. Most Fortune 100 IT shops, he added, are under cost pressure and prone not to take unnecessary risks with mission critical servers that are likely to move at a much slower pace.
Over the next two years, he expects a lot of testing by users and "assuming that initial testing proves benefits in these areas, expect larger IT shops to migrate towards multi-core, chipset servers over the next three to five years," says Archibald.