New hardware designs, including server chips that combine multiple CPUs on a single processor, will challenge the way enterprise software vendors price their products and could eventually force them to adopt new pricing schemes altogether, according to analysts.
Most large businesses pay for their databases and enterprise applications on a per-processor basis, coughing up an annual license fee based on how many chips they use to run their software. The model is simpler than pricing on a per-user basis because vendors and customers don’t have to constantly keep track of how many employees, partners and customers are using their software.
Server chips that combine two or more cores on a single processor could throw a wrench in the works, making it hard for vendors to price their products in a uniform manner. Further out, nascent technologies like grid computing pose even greater challenges and eventually could force vendors to adopt “utility-based” pricing schemes that charge customers based on how much use they get out of their software.
“The CPU-based licensing approach has got its limitations and it’s only a matter of time before some fundamental changes are going to take place,” said Albert Pang, a research manager with IDC, in Framingham, Mass.
IBM Corp. already sells Power4 chips with two cores, which can offer better performance for customers while helping IBM keep production costs lower. In the case of IBM its pricing is fairly straightforward: It charges customers as if they were buying two separate processors, and software vendors price their products accordingly.
But Sun, in a move aimed partly at making its chips more competitive, has said it will price dual-core UltraSPARC IV chips planned for release next year as if they were single processors, and it hopes that software vendors will follow suit. Further out, Sun has said it hopes to combine as many as eight CPUs on a single chip, raising questions about what constitutes a single processor from a licensing perspective.
“It’s safe to say that not just Sun but also Intel and IBM have this challenge of defining what is a processor,” said Andy Ingram, vice-president of marketing with Sun’s processor and network products group. “Chip dies can have multiple cores and multiple threads, so what constitutes one processor?”
Intel is adding to the mix with a technology called hyperthreading, which allows a single chip to act somewhat like two processors when running specially written software. Analysts have said that hyperthreading can boost the performance of software applications by around 30 per cent compared to Intel’s standard server chips, further muddying the issue of how much use customers with different chip architectures get out of their software.
To date, software vendors have made little effort to differentiate among processor types. Customers who run applications on Intel’s 32-bit Xeon processors, for example, typically pay the same price per processor for their software as customers using 64-bit RISC chips from the likes of Sun, IBM and Hewlett-Packard Co., even though those chips can offer greater performance.
Oracle tried to address the issue three years ago with a new pricing scheme which sought, in part, to distinguish between Intel- and RISC-based servers. The resulting “universal power unit” pricing model was almost universally reviled. Enterprise customers complained that their software had become more expensive and Oracle quickly abandoned the system.
“Even though people agreed that the Unix boxes were more powerful than the Intel servers, at the end of the day they didn’t want to pay more for that additional throughput,” said Jacqueline Woods, vice president in charge of Oracle’s global pricing and licensing strategy.
But with Oracle and some other vendors encouraging customers more than ever to run software on clusters of Intel-based servers, promising lower purchase costs compared to traditional Unix machines, they may be forced to address the issue once more. If a significant proportion of customers embrace server clustering, software providers could come under pressure to distinguish among the various processor architectures with their software pricing, Woods acknowledged.
“At a time when we see that happening we will evaluate it, but I’m not seeing that currently and at this point that’s not the plan,” she said. “The plan is to continue with the same pricing across all the hardware platforms.”
IBM, which also charges the same price for its software regardless of the hardware it runs on, also sees no reason at present to alter its pricing structure. But a representative declined to discuss any impact that emerging technologies may have, conceding only that “it’s an adapting industry.”
Other hardware developments on the horizon could pose even greater challenges to traditional pricing methods, making the issue of processor design seem relatively minor. For example, Sun, HP and IBM are all promoting the concept of “grid computing,” in which a datacenter could one day be managed as if it were a single computer, with administrators able to allocate processing power and storage to applications on an as-needed basis.
In such an environment, where computing resources are tapped like a utility, pricing software on a per-processor basis becomes essentially meaningless, analysts said.
“Current software pricing models are going to be totally broken by utility computing,” said Gordon Haff, an analyst with Illuminata Inc. in Nashua, N.H. “This (question of) what counts as a processor is just an additional wrinkle.”
Servers sold on a “pay as you grow” basis also pose potential difficulties. Sun and HP have both recently offered customers the option of buying a server with, say, four processors inside, but paying for only two of the chips at the time of purchase. As business grows and demand for processing power increases, customers can pay to turn the additional processors on.
That’s fine for hardware vendors, but for software vendors it becomes difficult to keep track of how customers are using their software, Oracle’s Woods said. Hardware vendors will need to work more closely with their software partners to figure out how to keep track of when additional CPUs come online.
Some analysts said enterprise software pricing is on the verge of a significant transition. In the days of client-server computing, when it was relatively easy to know how many users were accessing an application, per-user pricing was common. With the rise of the Internet and Web-based computing, vendors switched to per-processor pricing as a way of dealing with unknown user populations.
The challenges posed by multicore chips and grid computing, as well as slimline blade servers and other evolving types of hardware, put the industry on the verge of another inflection point at which a new pricing model may have to be introduced, analysts said.
IDC’s Pang said per-processor pricing is likely to hold sway for another two to three years, in part because vendors are reluctant to switch to a model that may require them to monitor more closely the way customers are using their software.
“Clearly there’s resistance from vendors like PeopleSoft and SAP, but I think it’s just a matter of time for subscription pricing or another system less reliant on CPU-based pricing to take hold,” he said.
Illuminata’s Haff said changing hardware designs and the rise of grid computing could be “the final nail in the coffin that drives Oracle, kicking and screaming, into a more utility-style pricing model.”
But it’s not clear that that’s what customers want, Oracle’s Woods said. For customers, the main issue is having a pricing scheme that is fair, simple, and predictable. Customers don’t necessarily want a licensing scheme based on, say, transaction volume, that fluctuates according to the amount of business they do in a given quarter, she said.
“They want predictability to help them manage their cash flow. If I have 20 users in January and 75 in June and 150 the rest of the year and my average for the year was 125 users – that’s a lot to keep track of. If I buy a license for 200 I don’t even have to think about it.”
Finding an alternative pricing model that is relatively easy to implement will certainly be challenging. Vendors who offer outsourcing services already have some experience with subscription-based pricing, while others have experimented with charging customers based on transaction volume.
“But if it’s hard to define what’s a processor and what’s a thread, then defining what’s a transaction is even tougher,” said Nathan Brookwood, principal analyst with Insight 64, in Saratoga, Calif.
Complicating the matter further is the highly politicized issue of price-performance benchmarks – software and hardware vendors alike will support a pricing scheme that makes their products appear to be better value for money. Whether they will agree on a pricing mechanism that is also friendly to customers remains to be seen.