X
Tech

The software licensing shake-up

Dual-core Special Report Part II: The emergence of multi-core processors threatens to upset the traditional way applications are priced
Written by Colin Barker, Contributor

The software industry is readying itself for a shockwave — the effects of which will be felt far and wide. One analyst describes it as the 'perfect storm' that will 'engulf software industry business models'. The catalyst for this shift is the emergence of a new approach to chip engineering: multicore processors.

Although the technology has been with us for around three years — developed with processors like the IBM Power range, Sun's UltraSparc IV and HP PA-8800 — it has only really begun to have an effect in the marketplace during the past year. AMD and then Intel pushed the technology into the mainstream when they began to talk of manufacturing multicore processors in quantity.

These new processors consist of two or more cores on one chip. A core offers roughly the equivalent of from 60 to 90 percent of the power a single processor, so a dual-core processor can a boost of between 20 percent and 60 percent compared to its single-core equivalent. This amount of extra power in a computer 'for free' has come as a great boost for the systems business and analysts are confident that multicore will soon be the only way to go.

Gartner predicts that by 2008, Sun will integrate as many as 32 logical cores on one processor, offering a huge increase in performance. The analyst also claims that dual-core processors will appear in standard PCs by 2006 and by the following year all processor production will be multicore.

Such rapid adoption of the technology is good news for technology users but not necessarily for software vendors. With one or two exceptions, software vendors are looking at the emergence of multicore with suspicion, if not down-right hostility. The question as they see it, is a simple one. Is a dual-core chip, one processor, for licensing purposes, or two?

The three biggest suppliers of databases, Oracle, Microsoft and IBM, have each adopted different strategies towards dual-core.

Oracle at first took a hard line, insisting, back in April 2004, that a core of a processor would be treated as a processor so that users of a dual-core chip would be charged the same as users of two, separate, single-processor systems, or a two-processor system. In other words, the price would double.

Critics were quick to point out that this did not look like a tenable strategy. Is a dual-core system twice as powerful as a single processor machine? Microsoft, for example, estimates a performance increase of 35 percent to 40 percent over single-core for its own software running on a dual-core system. That's undoubtedly conservative and nowhere near the doubling in power that Oracle originally wanted to charge for.

In fairly quick order, Oracle adjusted its position and announced that it was going to charge for each core in a multi-core system as if it offered 75 percent to the power. So a dual-core system would cost 50 percent more than a single-core, putting Oracle in line with the industry expectations of real performance improvements from multi-core. In real terms, Oracle enterprise edition would have moved from $40,000 per processor to $80,000 for a dual-core but will now cost $60,000.

At the other end of the scale, Microsoft had already gone much further treating cores and processors in the same way. So Microsoft SQL Server Enterprise Edition would cost from around $20,000 to $40,000 per processor and this will stay the same, the company says, regardless of whether the processor has two, four or more cores.

The two strategies reflect the relative positions of the companies in the market. Oracle's main source of revenue is database software, while for Microsoft, databases are an area the company is trying to grow into.

In the middle of the two lies the third major database vendor, IBM. True to form, IBM's strategy on software pricing for dual-core and multicore systems ploughs a furrow straight down the middle. "The value and real-world performance of dual core chips is variable and should be priced accordingly" a company spokesperson claimed in a recent statement.

The key factor in pricing should be the customer's ability to exploit the performance improvements according to IBM. "For third-generation dual core chips such as IBM's Power4, Power4+, Power5, where customers and partners have cultivated the expertise to fully exploit the performance of dual core on that particular platform, customers pay for two processors worth of performance".

The statement is curiously worded since it appears to say that because customers have, though their own efforts got smarter at exploiting the power of multi-core chips they should pay full price for them — scant reward for ingenuity one might think, but in line with Oracle's strategy.

However in the case of the first x86 dual core chips, most customers will not immediately use the chips' potential according to IBM and so it will "license x86 dual-core (and lower-end IBM OpenPower dual core chips used in Linux servers) as if customers were using the equivalent of only one processor license".

There is no single strategy covering software pricing on dual-core machines. Microsoft's is most generous, Oracle's most restrictive and IBM tries to read a middle course. The rest of the software industry is struggling with the same issue.

In October 2004 analyst Gartner argued that "ISVs need real performance metrics from hardware vendors to understand how to adjust pricing based on realised performance". Based on this "some vendors may choose to have special pricing to fairly account for multicore designs" which is the Oracle and IBM (on high-end Power systems) approach while "some may choose to price per-processor", the Microsoft and low-end IBM approach.

Gartner predicted that "by year-end 2005 at least three major ISVs will adjust their pricing so that per-processor pricing does not equal per-core pricing". Up to recently, Microsoft was the sole, major ISV to jump to per-processor pricing with IBM taking a half-step. In a bold move, the virtual infrastructure software supplier, VMware, joined fray. The company announced that its entire range of server virtualisation products would support dual-core processors and server products would be priced on a per-processor basis.

Virtual machines virtualise hardware so that multiple operating systems can be partitioned and dynamically scale the power available to any of the hardware resources. The more hardware available to use, the more partitions and the greater the flexibility that the virtual machines have, so virtualisation and multicore are two technologies that fit naturally together.

"Given the tremendous processing capabilities of dual-core systems, they are a sweet-spot for virtualisation," said Jeffrey Engelmann, VMware's executive vice-president of marketing. "Anyone deploying a dual-core system should implement... virtual infrastructure to harness the full potential of this enhanced capability."

VMware already supports dual-core system with GSX Server 3.2 and the client software VMware Workstation and ACE while the next releases of ESX Server and VirtualCenter will also support dual-core. It is perhaps not surprising that VMware should move quickly to embrace multi-core processors.

But while VMware's decision to welcome multicore was perhaps predictable given the natural fit between virtualisation and multi-core it was also helped by Microsoft's adoption of multicore. VMware's main competition comes from Microsoft Virtual Server 2005, as well as from the open source firm XenSource.

Multicore processors and virtualisation are two strands of development running through the marker which will contribute to what James Governor, of analyst firm Red Monk, describes as "the perfect storm set to engulf software". He claims the technology's impact software pricing will continue to gain momentum and anyone negotiating with application vendors needs to consider its impact. "They should closely scrutinise attempts by suppliers to lock them into long term deals using traditional pricing models, such as per processor pricing. End users might even consider get-out clauses in deals, to prevent long-term lock in," claims Governor.

The disruptive effect of multicore technology may seem intimidating but in the long term it should give users much more scope to negotiate with suppliers according to Governor. "Software vendors might not like it", says Governor, "but they are the masters of sneaky contractual clauses, so a dose of their own cod liver oil might not go amiss."

Editorial standards