Bus Standards: A New Bottom-Up Model

Intel is the company that everybody loves to hate—even though we admire its chutzpah and success. But the company has done some things right, including the art of specsmanship and market acceptance for open bus/interconnect standards. Intel's management, and more specifically its key technical folk, deserves kudos for what they've accomplished.

Almost single-handedly Intel has changed the way we do system and bus specs, and how we meld them into mainstream design. Intel's developers had key technical help and support from other industry vendors, but it was Intel that drove many efforts and made things happen.

The new Intel standards acceptance model is a bottom-up, specifications-driven activity. The idea is simple:

• Create a readable, usable specification
• Get the technical folk involved in tuning and spreading the spec with its technology
• Build a base from Intel's acceptance of the spec and from a bottom-up recruitment of companies to use and deploy the specification technology.

The impetus to spread the use and deployment of the specified technology is a technical sell, one that goes bottom-up—from the technical folk working with and reading the specification, upwards to their management—level by level.

Before you dismiss this model, consider PCI, as it moved from the its PC base to mainstream engineering. Yes, PCI's low silicon costs and relatively high performance helped, but it was also an understandable specification, relatively speaking. If you don't believe it, compare the PCI spec to an IEEE spec like 1394. It's an eye-opener.

And today PCI is probably the most successful bus, even surpassing the PC's AT bus and variations. Unlike any other bus, PCI now exists at four different implementation levels: system bus (CompactPCI, PCI/ISA), I/O bus (PCI), mezzanine bus (PMC, PC-MIP, PC/104-Plus, etc.), and intra-crate connection (National Instrument's MXI-3, TI's serialized PCI, SBS Technologies Dynamic Data, PLX's Seabring Ring, and others).

Intel used this spec-driven model with a number of different bus and system architecture specifications. These include I**2O, the Intelligent I/O Initiative; Universal Serial Bus (USB), the mainstream low-end peripheral connection for PCs and laptops; and InfiniBand, the pseudo-serial, crossbar-based switch fabric for intra-system connections. This very same model was also used for the Virtual Interface Architecture (VI), a low-overhead message-passing model for SANs (jointly specified by Compaq, Intel, and Microsoft).

Take a look at these specs. They might not be an easy read, the concepts may be complex, there may be some missing technical detail, but the specs are indeed readable. Technical people can read them and from the specs they can help to promote and move these technologies from proposals to concrete mainstream technologies and products. This is happening now: USB is a standard; and InfinBand is on its way to becoming a mainstream standard. Of the three Intel driven specs mentioned, only one may not be on the path to the winner's circle—I²O is stalled, waiting for its defined integration with the emerging InfiniBand specification.

Other vendors, some with competing technologies, also use this bottom-up model to drive acceptance of their specs. Technologies that used this spec-driven, bottom-up path include PCI-X, a high-performance extension of PCI (pushed by the Big Three system vendors: Compaq, HP, IBM); RapidIO, a lower level switch fabric for chips and boards; and FutureIO, the erstwhile competitor to Intel's NGIO (now integrated with NIGO as InfiniBand.)

This all sounds good: readable specs, developers driving standards acceptance, good technologies open for use. So what's the problem? Where's the pothole on this highway to success?

Actually, the problem is that the bottom-up strategy has been so successful. Unfortunately, it may take too long to move a technology up from the designers to executive suite acceptance. For in today's markets there may not be enough time to percolate up.

Specsmanship is good and it works. But it may work best as part of a dual pincer movement:

1. A readable, bottom-up specification thrust
2. A high-level techno-marketing thrust for upper management.

Intel did both with PCI. So if you're working for a new standard, push the readable specs, but don't forget the high-level technical marketing.