Special Report

Like it or not, we live in a market-driven economy. Discovering new and exciting techniques, technologies, and potential products is tantalizing, but the ultimate success or failure of a product, a product line, and the organizations that promote such ventures is driven by market factors.

Product life cycles, time-to-market requirements, proprietary vs. open architecture considerations, and operating gross margins mean more to product success than do the technical jargon that characterizes a potential product. Socket 7, Socket 1, and Socket 2 are much more than technical connectivity issues. They are components of the business strategy by which Intel creates economic chaos for Pentium clone manufacturers such as AMD and Cyrix.

Will CompactPCI (CPCI) become another Futurebus+ or a rising star that will steal market share from VME? Will CPCI have a limited life cycle and be challenged by technologies now entering the desktop marketplace, or will it become the next VME and develop a legacy lasting many decades? In order to make sense of these questions, we must examine the economic and market forces that favor or retard CPCI's emergence.

How Did CompactPCI get its Name?
It began in a small room. Names were offered up: Industry PCI, Rugged PCI, and Futurebus PCI, among others. The attendees were members of the PCI Industrial Computer Manufacturers Group (PIC-MG) who had come together to define a new specification for an industrial PC architecture that would include the electrical specifications of the Intel-developed Peripheral Component Interconnect (PCI) specification, and European mechanical specifications that were utilized by VME.

The attendees were aware of the opportunity to exploit the momentum of the recently developed PCI technology, and were acutely aware that the edge connectors and other features that characterized desktop PCI were inadequate to address industrial needs.

Finally, as the story goes, the name CompactPCI was put on the table and the membership agreed that name sounded cool (notwithstanding the fact that the first specification release contained the name Rugged PCI), and a product concept was created and named. Should the movie Son of PCI be made, Steven Spielberg would no doubt make the story more romantic, but according to Ziatech's Jim Medieros, a pioneer of CompactPCI, no significance should be attributed to the name beyond the fact that it sounded good.

Three years later, CompactPCI (or CPCI, to the displeasure of PIC-MG and the convenience of editors everywhere), has failed to live up to its initial hype. Now positioned more realistically (PIC-MG probably made a mistake when they predicted the end of VME), the market shows much promise but little in the way of revenues.

The marketplace is a fickle patron and the many glowing CPCI reports have turned to complaints and criticisms. Memories of Futurebus+ dance in the heads of more than a few industrialists, many of whom are reconsidering their original enthusiasm for CPCI.

What, then, are the facts surrounding CPCI, and how do newer emerging technologies-as well as market and economic factors-relate to the future of CPCI? Pushing aside the hype from both sides, we hope to take a realistic view of CPCI's technology, current and projected market, and the technological and economic forces that drive or retard its acceptance.

Beginning With the Standard PCI Specification
The concept of Peripheral Component Interconnect has grown from its original meaning as characterized in the original PCI bus specification to include some 13 different architectures and form factors. Since CPCI is predicated on the electrical characteristics of the original PCI specification, beginning there would be appropriate.

First realize that PCI and CPCI are different products, enjoy different markets, and are driven by different economic and market forces.

PCI was launched in the second half of 1994 and generated $40 million in embedded revenues that year. By 1996, revenues had grown to over $230 million, and 1997 embedded revenues exceeded $320 million. Embedded CompactPCI, notwithstanding grandiose predictions, went from zip to $5.1 million in 1996, reaching $18.2 million in 1997. Venture Development Corp. has been criticized for conservatively projecting 1997 revenues of $22.5 million (most other predictions were in the $50 million to $200 million range) and 1998 revenues of $61 million.

So why has PCI prospered within the embedded marketplace while CPCI has languished? Strangely, PCI has found life within the technology it was to have destroyed.

PCI continues to live up to and surpass the expectations thrust upon it by its progenitors in 1994. PCI's success in the embedded marketplace came not as the dragon-slayer of the monster VME, but rather as a result of its adoption by VME, enabling it to serve market requirements as a stand-alone bus architecture, as a local bus for VME applications, and as a local bus for embedded PC architecture applications. Rather than becoming the technology to replace VME, PCI has become, in a sense, more of a savior than a competitor.

Scaled back to use as a local bus, designers can effectively utilize its burst-mode bandwidth as a means of controlling a limited and well-defined set of I/O requirements.

The PCI specification was fathered by Intel in 1992 as a scheme for motherboard connection between CPU and high-speed peripherals, and handed off to the PCI Special Interest Group (PICSIG) to be raised and nourished. A son of many fathers, PCI's growth was enhanced after its adoption by the monster VME.

PCI's success was ordained as the inheritor of the PC architecture legacy, which offers distinct advantages to many embedded applications, including extensive hardware and software tool support (including C and C++), reduced software development expense, available off-the-shelf software (software transferability), and the ability to operate with inexpensive peripheral components (low cost ICs). Program developments made on the PC desktop, for example, are directly transferable to embedded systems operations, thereby saving costly development work and enhancing time-to-market considerations.

PCI offers high density and high integration, quality connectors, low cost, and a broad range of inexpensive multi-functional chips. With software development representing up to 90% of development costs on many projects, software transferability becomes an important design consideration.

PCI, however, is not without flaws. The specification, originally created for onboard local buses with a few add-in cards, was not intended for passive backplanes, sophisticated interrupt capability, nor multiprocessor support. The PCI bus, like its sibling CompactPCI, is slot-limited and must rely upon PCI-PCI bridges to increase the number of expansion boards in the passive backplane.

Many analysts, Venture Develop-ment Corp. included, have felt that the inability of PCI to address the ISA legacy (four PCI interrupts vs. 15 ISA interrupts plus DMA) would divide the market for embedded PC architecture designs, thereby reducing the market for PCI. The market hasn't responded in the manner that was anticipated. Many embedded PC boards in 1997 contained both ISA and PCI connectors. Many PC designs contain multiple PCI local buses (each with a PCI-defined expansion capability) and a single ISA bus to address ISA legacy interfaces.

There's a comedy routine about a father attempting to discipline his son by telling him "I brought you into this world, and I can take you out of it." Imagine the father is Intel (and its relationship with Microsoft) and the son is PCI, and the story gets interesting.

Both companies share the view that the days of having to open a PC to enhance functionality via add-in cards are nearly over. So the future of desktop computing seems certain to move away from parallel add-in buses to a serial interface for adding and connecting peripherals. USB and IEEE-1394 (Firewire) are lower-cost serial interfaces that require fewer connector pins and can operate on interface silicon that can be embedded within the CPU. Both USB and 1394 are today's technology (both will be ported on Windows NT v. 5.0, due for release in late 1998) and offer auto-configuration software and hot plug hardware. Being serial buses, USB and 1394 can scale in frequency into the Gbit/s range.

USB and 1394 won't eliminate the PCI bus in the future. The PCI bus will remain a local bus technology and require only two I/O chips: PCI-to-USB and PCI-to-1394.

THE CompactPCI STORY
CPCI has been an enigma ever since it burst onto the embedded merchant board market. Originally promising hot-swap, live insertion, and multiprocessor capabilities at a low cost, the ruggedized CPCI specification 2.1 has generated significant levels of enthusiasm within the telecommunications and industrial automation industries.

Computer designers have embraced PC-based embedded technology because of its extensive software development environment, its software development tools, and the fact that Microsoft-based software and development tools are significantly less expensive than Unix-based software and development tools.

Whereas ISA and PCI boards have found their way into only a limited number of telecom and industrial automation boards-due to the poor reliability of the mechanical slot card format-CPCI has solved these problems by adopting the IEEE-1101.1 mechanical specifications that were originally used on the early VMEbus specs.

In addition, the CPCI form factor for a 6U board leaves an additional 220-pin connector which can be used for a variety of purposes, thereby providing additional flexibility for embedded designs.

For the good or bad, CPCI designs are based on Windows NT software and Intel microprocessors. Windows NT is nondeterministic and its NT 4.0 configuration is years from being stable; NT 5.0 is expected out by the end of 1998, and rest assured that bugs are virtually guaranteed.

The dependence on PC market silicon illustrates CPCI's vulnerability to the whims of the chip and software vendors. Windows NT 5.0 will provide drivers to support IEEE-1394 and USB, serial buses that are expected to dominate the desktop by the beginning of the millenium. Since CPCI is PC architecture-dependent, and since the PC is heading towards the very-high speed serial bus, the question regarding what effect, if any, USB and IEEE-1394 will have upon CPCI is uncertain.

The development of bridge technologies by chip manufacturers is having an impact on current and future designs. A cadre of PCI-to-PCI bridge developers have provided a means for PCI slot expansions, notwithstanding a host of associated programming difficulties for the users of such technologies. Tundra Semiconductor has had great success with their Universe chip that permits either direction VME/PCI interoperation. The net effect is to bring PCI into the VME world, thereby increasing the total market for PCI technology.

Tundra's recently announced virtual PCI chip (named vPCI) may have an effect upon the use of CPCI in embedded designs. This technology allows a CPCI motherboard to control a full 21-slot, multiprocessor capable, full interrupt-capability VME backplane that contains a local PCI bus interface on each VME card. The vPCI system is designed such that the VME backplane is completely transparent to the CPCI-operated software. In such, desktop-developed software can be directly used to control a VME backplane as if the system was a PCI or CPCI design.

Whether or not vPCI will address VME legacy is unclear; however, this development represents a clear example of how CPCI technology might inadvertently serve VME expansion (as did PCI), based upon market forces (the large VME legacy that could now use desktop PC software).

Telecommunications: The Logical Battlefield for CPCI
CompactPCI, although currently showing more promise than revenues, is not to be taken lightly. CPCI, by nature, lends itself to the packet-switching fabric that characterizes modern telecommunications. Telecom engineers like the CPCI chassis that is designed to accommodate a large number of rear-connect phone lines, allowing unrestricted card insertion and front-end card swapping. They also like the CPCI form factor for a 6U board that leaves an additional 220-pin connector, which they can use for a variety of purposes.

Vendors are drawn to CompactPCI for several distinct reasons. First, service providers, the consumers of telecommunications equipment, are committed to vendors who offer an open architecture (so as not to become enslaved to a single manufacturer). This situation places a demand on vendors who can offer an architecture that will run a variety of common software. Second, time-to-market considerations are dominated by software development more so than hardware issues. Lastly, the PC architecture possesses the richest software development environment in the world.

Computer designers have embraced PC-based embedded technology because of its extensive software development environment, its software development tools, and the fact that Microsoft-based software and development tools are significantly less expensive than Unix-based software and development tools.

A principal force favoring CPCI for telecommunications applications is its ability to support Windows NT, the favored development platform for Advanced Intelligent Network (AIN) software developers. The Intelligent Peripheral (IP) AIN platform requires enhanced services to be customized for each subscriber. Windows NT lends itself nicely to this task. It was assumed that the requirements of the Service Control Point (SCP), the AIN database library, would remain within the Solaris domain, because of its reliability, scalability, and robustness.

Surprise! Siemens has committed to using Windows NT and DGM&S software on its SCP. This move might send shock waves across the industry, as others move to NT. It's no secret within the AIN telecom industry: support NT or be left out.

Major telecom vendors have indicated that they intend to use CPCI on their IP platforms, although no substantial orders have been placed, so it's sort of a tease. Alcatel, Nortel, Siemens, Lucent, and many others are evaluating competitive boards to integrate into their own systems. Yet very little money, if any, has changed hands. Certainly, there exists a strong demand by the telcos for open architecture systems, so that certain vendors don't entrap them, as they have in the past.

CPCI's Problems
The VME marketplace broadly supports a variety of microprocessors and operating systems. Some 24 companies control 75% of the total market, thereby enhancing niche markets and competitive opportunities. The CPCI marketplace, conversely, is dominated by such stalwarts as Motorola, Sun Microsystems, Force Computers (supported by Solectron, Inc.), and Ziatech (supported by SCI). No longer the inexpensive alternative to VME, the only current market of significance is within the telecommunications area, and purchasing decisions among the large telecom vendors will be based on manufacturing acumen and cost containment. Gross margins have been severely slashed and there is little room for innovative niche markets to open. The smaller companies pursuing CPCI will be addressing traditional ISA markets in search of lesser payoffs.

In an attempt not to repeat mistakes of the past (remember Futurebus+?), CompactPCI is struggling to allow its specification to be fluid enough to accommodate emerging applications, yet specific enough to serve as a real specification that will lead to an emerging, rather than fractionating, industry. This balance is difficult to achieve.

For example, more than 25 legal backplanes are possible under CPCI rev. 2.0. Whereas the looseness of the specification can lead to more profitable custom work, it acts as a deterrent to growth. For example, Motorola is attempting to drive their version of the CPCI backplane to de facto status. Sun and Force are doing the same. Conceivably, many CPCI systems will be installed that are incompatible with each other.

R. Brough Turner, chief technology officer at Natural Microsystems, Inc., has made clear his company's commitment to addressing the wealth of PC software that is available. He has stated that he will utilize CompactPCI, PCs, or potentially any other platform that will function within the wide range of PC software.

To no one's surprise, Intel has announced its successor to PCI, to address emerging ultra-high bandwidth applications. This announcement should not affect CPCI for a long time. The driving force behind CPCI is its ability to use desktop-developed software. However, should the CPCI market remain a near-commodity status environment dominated by a few large players, the remaining members of the highly talented embedded industry will find another technology with which to challenge CPCI.

Jerome Krasner is group manager of telecom and embedded systems for Venture Development Corp. He received BS and MS degrees from Washington University (St. Louis, MO), a PhD in medical physiology from Boston University, and an MBA from Nichols College (Dudley, MA). He writes frequently on topics of technology markets and medical electronics.