The three companies are promoting an input-output serve architecture, called Future I/O, which was detailed today at a conference here. Chipmaker Intel and a host of other companies are backing a different standard; however, its three largest customers are Hewlett-Packard, Compaq, and IBM.
But Intel is now one of the 40 companies that has signed on to evaluate the Future I/O technology, executives said at the meeting. Intel's presence does not, however, commit the chipmaker to the competing technology.
At stake in the dispute is the way that components like network adapters are plugged into servers for years to come--a fundamental part of the design of expensive servers. Each side claims to have the advantage in performance.
But more importantly, power and profit hang in the balance. Future I/O will allow these companies to control the future development of servers and even bring homegrown supercomputer technology to the masses, according to many. Intel's control over the effort, by contrast, allows companies with low R&D budgets like Dell to undercut these companies, Future I/O executives have said.
"We're in almost daily conversations at the highest levels," said Karl Walker, vice president of technology development at Compaq. Merging the server company's Future I/O plan with Intel's competing Next Generation Input/Output (NGIO) is "our exact plan. We think it will be resolved before [the technology] hits the market," Walker told News.com.
One significant step that brings the two camps closer together is that the Future I/O camp has changed its royalty plan so that those who wish to use Future I/O in their products must pay an annual fee instead of a per-unit royalty, said Tom Bradicich, director of IBM's Netfinity server architecture and technology. Companies that don't sign up for that plan would have to use the relatively cumbersome approach of licensing the Future I/O technology.
Despite the olive branch Walker extended, merging the two efforts won't be a simple task, and the Future I/O group clearly has reservations about NGIO, partly because concerns that NGIO is too slow to meet customer needs and partly because companies don't like the legal ramifications of joining the NGIO effort.
Intel has representatives at the Future I/O conference, but said that the company has not joined the FIO effort, said Intel spokesman Bill Kircos. Intel will go ahead with its own conference on NGIO and other technologies at its Developer's Forum beginning February 22.
Intel couldn't yet comment on the specifics of the new FIO payment scheme, but added, "We'd like to see the core specification be open and free," Kircos said.
FIO may be faster, but NGIO is coming out sooner, with products due in 2000. FIO advocates said today that they expect the standard to be ratified by the end of 1999, with products shipping in the first quarter of 2001.
NGIO too slow?
NGIO, which can transfer data at 125 megabytes per second, is actually slower than the Peripheral Component Interconnect (PCI) bus on today's desktop computers, Walker said. "We're a little concerned that they're undershooting the target," he said.
Future I/O will allow data to be transferred 16 times faster than NGIO, he said. Specifically, Future I/O allows 1 gigabyte per second in two directions, for a total peak bandwidth of 2 GB/sec, he said.
"I see no reason to invest in a technology that is equivalent to Gigabit Ethernet in performance," said Bill Huber, director of engineering at 3Com. 3Com, a major network adapters manufacturer, has firmly sided with the Future I/O effort and along with HP, Compaq, IBM, and Adaptec is now one of the five "promoters" of the technology.
Adaptec, another adapter manufacturer, is more cagey about which way it's leaning. "We're eager to see a single standard emerge," said Bob Selinger, chief technologist at Adaptec.
Another issue is that Future I/O is designed to work with several different chips, and Future I/O advocates were concerned that NGIO would work only with Intel chips. HP, Compaq, and IBM want to use Future I/O for their Intel-based server lines and machines using other chips--specifically, HP's PA-RISC chips, Compaq's Alpha chips, and IBM's PowerPC chips--executives said.
This versatility could be key to one of the points often recited by Future I/O backers, that the standard is a foundation that allows companies to build superior products. NGIO doesn't have the what's needed for companies to differentiate their products, Walker said.
But technical considerations are in some ways secondary to the legal obstacles. To participate in NGIO, a company must grant Intel a complete license to that company's intellectual property. "It gives Intel any piece of intellectual property Intel chooses," said Martin Whittaker, manager of research and development at HP.
Moreover, companies must sign away their IP before being able to evaluate whether they would want to back NGIO, Huber said. And it's not a two-way street where companies would get equally open-ended license to Intel's intellectual property.
Intel has argued, however, that charging money for use of an interconnection technology will doom it to failure.
Walker said the intellectual property rules for the Future I/O effort will encourage companies to share their "crown jewels." Companies that contributes to the effort will get a share of the royalties that will be charged for users of the technology who don't share their intellectual property, Walker said.
Bradicich said it's not yet clear what fee will be charged to participate in the Future I/O effort.
However, he emphasized that royalties will be merely "chump change," and that the real profits from Future I/O will come from a technology that allows server makers to make products superior to competitors' hardware--a contrast to the indistinguishable products often found in lower-end computing products today.
Future I/O and NGIO are fundamentally similar. They both use a "switched fabric" or "channel-based" architecture, essentially a miniature network inside the computer that lets a computer establish dedicated connections between the CPU and peripherals such as network adapters or disk storage systems.
The new architecture is an improvement on today's prevailing technology, the PCI bus, in which several devices share the same data pathway. Switched fabric designs not only offer faster speeds, but also more robust computers: With the new design, it's easier to track down errant equipment, it's harder for devices to scramble the system's memory, and it's easier to switch out devices without having to shut down the computer.
Intel's NGIO effort has the jump on Future I/O. Intel began NGIO more than two years ago, but Future I/O began when IBM, Compaq, and HP started discussing it early in 1998. The groundwork for Future I/O was laid when the triumvirate began work on PCI-X, an extension the companies initially developed without Intel's help.
Now Intel is a part of the PCI-X specification development, and Walker said he hopes a similar collaboration will emerge with Future I/O and NGIO. Intel technical personnel were attending the conference, he said.
Also in attendance were staff from Microsoft, Santa Cruz Operation, and Novell, three companies that sell operating systems. For Future I/O to become adopted, operating systems must work with the new hardware.
The first hard details on Future I/O emerged today at the Future I/O Developers Forum, which Walker described as "the first coming-out party" for Future I/O. Microsoft joined IBM, Compaq, HP, Adaptec, and 3Com in giving presentations to an audience of more than 220.
One company that has signed on to evaluate Future I/O is Poseidon, a privately owned San Jose start-up that has developed a switched fabric technology of its own. Future I/O could mean Poseidon might supply chips to the server vendors, said Chief Executive Rick Shriner. Shriner said he was impressed with the turnout at the conference. "There are a lot of architects sitting in that audience," he noted.