Sun Microsystems is urging that the negotiations between two high-profile camps attempting to define the next internal architecture for server systems be finished by August, or the negotiations be dropped altogether, according to a source familiar with the negotiations.
At stake in the battle of the
|Key dates in the server war|
The struggle to set the next-generation server architecture standard has been going on for months.
|Nov. 11, 1998||Intel releases first details about NGIO.|
|Jan. 7, 1999||Intel announces that Dell, Sun, and others have joined the NGIO Forum.|
|Jan. 13, 1999||Compaq, IBM, and Hewlett-Packard announce Future I/O, without Intel support.|
|Feb. 11, 1999||The two camps discuss plans to resolve the dispute; a Future I/O representative predicts resolution before the technology hits the market.|
|May 24, 1999||Legal obstacles are resolved, but technical issues block a merger. Future I/O announces its technology will use the same communication method as the Internet.|
|July 20, 1999||Version 1.0 of NGIO specification is announced.|
One thing all parties agree on is that they need a successor to the current technology, the PCI bus, which companies say isn't fast enough and is responsible for many computer crashes.
Sun's deadline tactic puts pressure on IBM, Compaq Computer, and Hewlett-Packard, which advocate a system called Future I/O. That system is pitted against Next-Generation Input/Output, or NGIO, backed by Intel, Sun, and Dell Computer.
NGIO representatives wouldn't comment on the negotiations. But in an earlier interview, Sun executive Charles Andres said the NGIO Forum "didn't want to delay the specification of NGIO because of a merge that may or may not happen. We have to keep moving on NGIO regardless."
Future I/O and NGIO representatives said talks continue on a daily basis and that both sides agree they would like to come up with a single specification.
"We believe the right answer is to get a single specification, provided that the specification meets the needs of our customers," said Martin Whittaker of HP's server research and development group.
"I remain cautiously optimistic" that the two groups would join the standards, Whittaker said. "I would be disappointed if folks were setting arbitrary deadlines."
The two camps have been negotiating to merge the two standards for months, trying to avoid the expense and difficulty of supporting two competing standards. While the two camps have resolved their differences over issues such as intellectual property and licensing fees, technology issues still are holding up a merger, representatives of the camps have said.
The NGIO Forum wouldn't comment on the deadline issue but, in a statement, said the two camps are continuing to talk: "We believe one spec is better for the industry."
NGIO, which initially was developed by Intel, is simpler but slower than Future I/O. Future I/O advocates say NGIO isn't fast enough to bother with, but NGIO advocates say Future I/O is too expensive to build. Though NGIO is slower, a higher-speed "fat pipes" version of NGIO is under way, and multiple NGIO connections can be added together to boost data transfer speeds.
Future I/O and NGIO are similar in that they use a miniature network called a "switched fabric" to handle the communications inside the computer, a method that offers higher speeds than the currently popular method of sending data to several devices sharing a data pathway called a bus.
The NGIO camp drew a line in the sand last week with the release of version 1.0 of the NGIO specification.
"The industry can start building implementations on the agreed specification," said Andres in an interview when the standard was released.
Regarding criticisms that NGIO isn't fast enough, Andres said the fat pipes version will be 10 to 100 times faster than the first version of NGIO. And while the current generation of NGIO is slower than Future I/O, it's also vastly simpler and cheaper, he said.
Intel is showing prototypes of NGIO chips, he added.
Even though version 1.0 of NGIO is out, there still is room for a merger with Future I/O, Andres said. "At the end of the day, it's possible we will back off some," he said.
The Future I/O camp viewed the version 1.0 announcement with skepticism. "It was an interesting move Intel made, declaring the NGIO specification, yet declaring six months later that version 2" would be announced, Whittaker said. "To me they're intrinsically admitting that they're not done and they haven't met customer needs."