This is the third in a short series of articles that Techworld is publishing over the next few days about multi-core processing and its ramifications. Read the first article here.
If Chris Easterwood, IT services director at Southwire, a wire-making company, installs dual-core chips in his servers, he may get an unwelcome surprise: a licence fee increase from his database software vendor, Oracle.
For pricing purposes, Oracle is counting a dual-core chip as two chips. That means if Easterwood puts a dual-core chip in a socket previously occupied by a single-core chip, he expects that his database software licensing costs will roughly double. As a result, he's less likely to adopt the new technology, he says. Although Southwire had revenue of $2.2 billion last year, wire manufacturing isn't a high-margin business, and IT dollars are watched carefully. If software vendors consider dual-core systems as two processors, Easterwood says, "then the extra benefit we get out of dual-core goes away."
In fact, some industry experts expect a broadly negative reaction to any decision by software vendors to charge based on core.
"Unless software companies and the industry in general start rethinking the whole licensing thing, they're going to be in trouble, because it's going to stop people from implementing the technology," says John Abbott, chief analyst at The 451 Group.
But despite user concerns, software licensing issues haven't slowed down the chip vendors. Intel and AMD expect to install dual-core technology in more than two-thirds of the servers shipped by the end of next year. In time, nearly all servers will have multi-core chips.
"The industry is going to have to cope with multiple-core processors as being the standard," says Graham Lovell, senior director of x86 servers at Sun Microsystems, which produces servers based on AMD's Opteron chips.
But while all servers might soon have dual cores, software licensing will be far from uniform. Indeed, three of the biggest software vendors seem to be taking different approaches. IBM will make licensing decisions on a case-by-case basis, Oracle will count a dual-core chip as two chips, and Microsoft intends to treat a dual-core chip as a single chip.
IBM has told Power5 dual-core customers that those processors will be counted as two chips because users are getting "full benefit" of the processors -- in other words, performance two times that of a single core. In response to an e-mail question, IBM said it can do this because the Power5 is in its third generation of dual-core technology. But installing a single dual-core x86 chip isn't the same as putting in two chips.
Chip vendors say dual-core users can expect performance gains of about 1.3 to 1.8 times that of one CPU, largely depending on the application that's running. Because of that, IBM considers the dual-core x86 chips as a single processor for software licensing. But as the chips improve, IBM might alter its plan and count dual-core as two chips. An IBM spokesman says the company will "constantly evaluate" its licensing approach based on customers' reported chip performance.
Oracle's licensing policy is similar to IBM's for its Power5 dual-core processors. The company says that as the dual-core x86 processors advance, they will offer performance equal or better than two separate chips.
That's in part why Oracle views dual cores as two processors for licensing, says Jacqueline Woods, vice president of global licensing and pricing strategy. In making her point, Woods cites IBM's approach for software licensing on the Power5 chip. While dual-core might not deliver two times the performance today, she says, it will likely offer performance gains of double or more. "As the performance on those chips does improve, our pricing will stay the same because we are agnostic in that area," Woods says.
Several years ago, Oracle tried pricing based on performance, with its universal power unit. But as chip megahertz increased by leaps, so did software licensing prices, prompting some user complaints. Oracle dropped the power-unit idea, focusing instead on per-CPU pricing.
Woods says the company is unlikely to return to a licence formula based on chip performance. Oracle's licensing prices don't necessarily double, even if the cores do, she says, and the company offers discounts as the number of chips increase, in addition to other licence plans that are based, for instance, on numbers of users.
Paying for power
But for users who pay based on CPU usage, dual-core is just the beginning. Chips with four or more cores based on the x86 architecture, as well as in RISC-based systems, will be arriving as early as next year. Indeed, multi-core chips can get very large. For example, a network-attached computing appliance recently developed by Azul Systems has 24 cores per chip.
Yuri Aguiar, senior partner and chief technology officer at Ogilvy & Mather Worldwide is testing the Azul systems. He says licensing models will have to adapt to new technologies, lest software vendors risk defections to open-source platforms such as MySQL or PostgreSQL databases.
"Even if the uptake is slow in the beginning, many of these databases are proven and will be adopted by more people than we expect if licensing terms don't change quickly," says Aguiar.
Users such as John R. Dick, CIO and executive vice president at Financial Corporation, want the increased power of dual-core chips without having to pay what they consider a software pricing penalty.
Dick sees dual-core chips as a "scalability option" for his firm -- a way to increase performance. He says he isn't likely to take a two-processor server and replace it with one running a single dual-core to cut his software licensing charges. He simply wants the performance gain. "The pricing proposition vis-a-vis other alternatives in the market would, over time, drive you to a different product," he says.
The reality is that any licensing change by the vendors will have a negative impact for some IT group somewhere.
In the case of virtualisation, per-CPU pricing "works to our benefit if we are able to deploy several system images on a CPU," says Madge Meyer, executive vice president and head of global IT infrastructure services at financial services firm State Street. Dual-core chips may "create a negative impact on the cost of licensing if the software vendors decide to treat a dual-core CPU as two CPUs, and they will probably build a strong case for doing so," she says.
It could mean having to move away from counting physical CPUs to pricing based on the number of times a separate application is running on the chip. Such a move would provide software vendors with some additional revenue "but will likely not take away all of the benefits of virtualisation so as not to kill the golden goose," says Meyer.
It remains to be seen whether multi-core chips will fundamentally change how users pay for software. Some believe that software vendors may have to move toward an open-source pricing model, earning revenue from maintenance and support and not from per-CPU pricing.
Randy Roth, a partner at Corporate Contracts, a consulting and contract negotiation firm, says software pricing is making users think harder about what they're spending. "We have been seeing more companies willing to limit the amount of software that they will license, because of the difference in the perceived value in what they get out of the software and the increasing cost of running that software."
The per-CPU pricing model was adopted for its simplicity, but virtualisation and multi-core technology change that, says David Znidarsic, vice president of technology at Macrovision, which makes licensing management systems.
He says few software applications can deliver "linearly increased performance" as CPU capabilities increase. The result, says Znidarsic, will be increasing tension between "the end users who are wanting proportional value of the price increases and the software publishers who are challenging their engineering departments to deliver proportional value to the number of CPUs."