By Siobhan Kennedy
Networking giant 3Com Corp became the latest high profile vendor to lend its weight to the Future I/O server specification proposed by Compaq Computer Corp, IBM Corp and Hewlett Packard and on show this week at their developer’s forum in Monterey, California. The forum, which was attended by over 220 people from some 90 hi-tech companies, was designed as a platform to outline the merits of the architecture and provide details of its technical specification. Although an official announcement is expected on Tuesday, Karl Walker, VP of technical development at Compaq said 60 vendors have already signed up to support the spec. He said it was too early to name names but he added that the list read like a who’s who of the systems, adapter card and operating systems vendors. Microsoft has already gone public with its support of the architecture. Walker said the steering committee had places for nine companies. With the inclusion of 3Com this week, and Adaptec, who joined the group in January, that leaves four places for other vendors, which are currently under negotiation to be filled. The forum marked the first opportunity that IBM and it partners have had to substantiate their claims that Future I/O offers vendors the ability to differentiate their products in a way that the rival specification, NGIO (Next Generation Input/Output) from Intel Corp, can’t. Both the NGIO and Future I/O technologies are designed to do the same thing, that is, to govern how disk arrays, network cards, and other components plug into servers, thereby replacing the Peripheral Component Interconnect (PCI) technology used today. While both camps have based their specifications on switched fabric technology, Intel says the main sticking point preventing the two from working together is the issue of how to charge IHVs (independent hardware vendors) who want to use the specification in their products. Walker admits that the group will charge a nominal fee for the license, something in region of a few thousand dollars but he argues that vendors will get specification that is technically superior to NGIO.
Future I/O is faster than NGIO
First he points to performance issues. Walker said Intel’s NGIO allows data to be transferred at 1 Gigabit per second whereas Future I/O carries it sixteen times faster, with a total peak bandwidth of 2 Gigabytes. More importantly, he added that PCI-X, IBM and its partners’ interim architecture to succeed the standard PCI bus of today’s PCs before Future I/O takes over, runs eight times faster as NGIO. What’s the point of Intel delivering a new architecture that will already be slower than the existing PCI specification? It doesn’t make sense. he said. Moreover, he said Future I/O would be backward compatible with PCI-X so customers won’t have concerns when it comes to upgrading the machines to a different bus architecture. It’s the ability to maintain stability that Walker says is key and something which Future I/O, not Intel’s spec, can provide. Our prime directive is to give stability in the enterprise marketplace, he said, technological churn is happening much too fast. We upgrade chips every three months and people get stressed. But he added you can multiply that by a factor of 100 when it comes to the I/O system. The level of sensitivity of applications to a change in the I/O architecture is incredible, he added, changing PCs is a pain, but changing the interface that governs the connection to the Lan, the internet, to data everywhere, that’s incredible. Walker argues that Future I/O has been designed to guarantee maximum stability. Not only is it much faster than NGIO, and will therefore need upgrading far less frequently, the planned upgrade path is seamless and backward compatible. Moreover, he said that although the architecture would be developed with a set of common standards, it had been designed in such a way as to enable vendors to add compatible extensions on top of the bus so each vendor can focus on their core competencies, whether it’s fault tolerance, throughput or speed. But when Intel introduced its spec, 90% of the work had already been done, which gives manufacturers little or no chance of differentiating their products, he said. He also uses this argument to justify charging royalties. The reason we did it that way was to give companies an incentive to contribute the best technologies they could. We all share the intellectual property and everyone benefits, he said, but with Intel’s model, all the intellectual property belongs to Intel, so effectively you’re giving them your check book. Despite his criticisms, Walker reiterated that the HP- Compaq-IBM triumvirate were still eager to get Intel on board. It’s nobody’s intent to have this be a confusion factor for the customers. We’re in almost daily negotiations with Intel and I’m pretty sure sanity will prevail. But a spokesperson for Intel disagreed. It’s a question of timing, he said, we’re moving forward with NGIO, the specification is now two years old, Future I/O has just been announced. He added: Any joint movement will have to come from their camp.