BY WAYNE FISCHER

If you’re looking for an impressive balancing act, forget about the circus and go straight to embedded computing where open standards, cost-effectiveness, performance, and tele-phonystrength reliability must all come into play, and where new bus implementations are changing the name of the game.

Like high-wire artists, people across a broad spectrum of industries — not least among them telecommunications — are looking for ways to balance the low-cost, open-standards-based advantages of desktop PC technologies with the robustness, reliability, and modularity requirements of their critical embedded applications. The poor thermal features, inadequate shock and vibration characteristics, and difficult serviceability of PC-based equipment all have to be overcome. Yet, wouldn’t it be desirable if this could be accomplished without losing the benefits of using PC silicon and the standard PCI local bus architecture, and without sacrificing compatibility with thousands of PC software programs?

BROADCOM, FOR EXAMPLE
The experiences of Broadcom Corporation, a leading Californiabased supplier of integrated chips for the digital broadcast industry, provide an apt example of how companies are starting to crack this conundrum. Having recently developed integrated chip technology for high-end systems — aimed at providing highspeed cable-modem Internet connections to our homes — Broadcom needed a pilot reference platform that used highspeed PCI bus technology, the PC industry’s local microprocessor bus of choice.

First published by Intel in 1992, the PCI bus specification defines a highspeed, processor-independent bus for interconnecting high-bandwidth I/O components. The PCI bus is designed to last many microprocessor generations over a 10 year plus period, instead of being redesigned for every new microprocessor generation.

Up to this time, Broadcom, like many companies, had primarily used conventional PCI-based PCs for this type of task, and had written the software on top of the existing operating system — either Windows 95 or QNX (a UNIX kernel optimized for real-time). The importance of this particular application, however, called for a far more robust electromechanical platform than that provided by traditional PCs.

COMPACTPCI — A 3-WAY PARTNERSHIP
Broadcom’s solution has been to not use a generic PCI solution at all. Instead they have implemented PCI’s industrial computing extension, CompactPCI. Electrically and logically, CompactPCI CPUs and peripherals are the same as their standard PCI counterparts, and CompactPCI systems utilize identical silicon, firmware, and software as traditional PCI bus systems. Operating systems, drivers, and applications can’t detect any differences between the two buses because, fundamentally, there aren’t any. As a result, CompactPCI implements the PCI’s inherent high performance (up to 264 MBps peak throughput), which is crucial for multimedia, video, networking, and other high-bandwidth I/O applications.

They also inherit PCI’s ability to support Pentium, Pentium Pro, PowerPC, SPARC, MIPS, Alpha, and various specialty microprocessors, its ability to benefit from high-volume silicon production, and its plug-andplay capabilities. And as the silicon, firmware, and software are indistinguishable between CompactPCI and standard PCI, transforming a standard PCI card involves almost no redesign. Rather, it is merely a matter of a physical relayout.

There are, however, key enhancements in the form of additional mechanical features to improve the maintainability and reliability of a PCI system in a telecommunication or other critical industrial environment. CompactPCI, as defined in 1995 by the 300-member PCI Industrial Computer Manufacturers Group (PICMG), is essentially a three-way technology partnership of convenience, involving:

• PCI for the electrical, logical, and software layers.
• Eurocard form factor for the mechanical layer (19”rack).
• 2-mm Hard Metric (HM) connectors for the interconnect.

For Broadcom, the addition of the Eurocard form factor (identical to that specified in the tried and true VME standards) combined with 2-mm connectors have enabled the design of a much more robust and easier to maintain platform that still maintains all the performance, openness, and time-tomarket advantages of PCI.

To understand how these benefits are arrived at, we need to first delve deeper into PCI technology itself, more closely examine the ways CompactPCI extends this technology, and discuss the relative advantages of CompactPCI and existing VME bus technologies.

ENHANCING THE PCI ENVIRONMENT
CompactPCI derives from the best of PCI bus and many of the best of VME bus technologies. Figure 1 depicts a generic PCI system — typical of a conventional or industrial PCI-based PC. As we’ve seen, the CPU/cache/memory subsystem represents virtually any kind of microprocessor in use today, from Pentium to Alpha chips. The peripherals are generic in that they are not dependent on the architecture specifics of the CPU/cache/memory subsystem, which is isolated by a bridge from the local bus. A major appeal of this architecture, in conventional PC systems, is the way it has encouraged the development of lowcost, high-performance peripherals.

The CompactPCI architecture encapsulates and extends the existing PCI environment to meet the more demanding requirements of the telecommunications, industrial, and related markets. Up to seven CompactPCI slots can be bridged into a PCI bus segment (represented by the first, or system, slot), with multiple PCI bus segments implemented in a single monolithic backplane. This contrasts with the three slots that are typically supported by standard PCI bus technology.

RUGGEDIZED PACKAGING AND CONNECTIONS
The CompactPCI boards that utilize these slots adhere to Eurocard packaging standards to help add industrialclass reliability and maintainability to the PCI environment. These boards can be either 3U or double-high 6U boards. Eurocard features include extensive “board keying” capabilities (4,096 combinations) so that specific boards can only be plugged into specific slots if the application requires, card guides for solid rear backplane connector alignment, front panel retaining mechanisms, solder side covers to protect other components, injector/extractor handles, and EMC protection features to minimize electromagnetic interference.

In contrast to conventional and industrial PCs (standard, desktop form factor PCs adapted to meet some of the challenges of industrial computing), the connectors themselves are rugged pin and socket connectors, similar to those used by VME. Unlike VME, however, these are higher-density 2-mm Hard Metric connectors designed for use in telecom equipment (ANSI IEC-1076) and now widely used in industrial control applications as well. Compared to the edge connectors used in standard PCI cards, these pin and socket connectors provide faster propagation times, reduced reflection at the bus/connector interface, lower noise, better impedance matching, and higher mechanical reliability.

The aggregate advantage of these pin and socket connections, card guides, retaining mechanisms, and the like, is much improved support and durability for every board and board connection in the system compared to standard PCIbased PCs. Maintenance, repair, and upgrading are also significantly simplified.

PASSIVE VERSUS ACTIVE BACKPLANE
CompactPCI boards are inserted and removed from the front of the system, in further contrast to conventional and most industrial PCs. A PCI card (or ISA, or any other type of card) in a typical industrial PC is inserted and removed from the top of the motherboard, to which it is fastened only by its top corner — a relatively flimsy solution in a telecom or other industrial environment where the effects of shocks and vibrations have to be taken into account.

To compound the problem, in the active backplane configuration used by many industrial PCs, the motherboard acting as the mechanical support contains active components that are now subjected to the strains associated with supporting card connections. Replacing an active motherboard when any of these components fail is not a trivial task, often requiring complete system disassembly. CompactPCI implementations, on the other hand, use only a passive backplane configuration. As we’ve seen, the CompactPCI system controller is just one slot in the 8slot CompactPCI system — easy to pull out and replace when making repairs or when upgrading the system.

This easy maintenance applies to every board in the system. When CompactPCI boards are removed, there is no impact on the rear I/O connections favored in telecom environments, or on neighboring cards. Neither is there any internal cabling between CompactPCI peripheral boards that needs to be removed before a board can be replaced.

ELIMINATING HEAT STRESS
Another major advantage of Eurocard packaging relates to even airflow throughout a system. Heat buildup caused by uneven cooling is a major cause of failure due to warped boards, the resulting bad connections, and short-lived components. Active motherboard systems are particularly susceptible to these problems. Unfortunately, the conventional desktop chassis that houses most industrial PCs does nothing to mitigate uneven cooling. In fact, it makes it worse because the rear of the chassis is used to mount all the I/O cabling and there is no easy outlet for air in that direction. Nor can cooling air — blown up through the peripheral cards — be used because the motherboard gets in the way.

The Eurocard system avoids these blockages so that each board gets an even amount of air over its surface. Generally it’s a matter of mounting two or more fans at the bottom of the rack to force air upwards along all the boards.

COMPACTPCI AND VME
In extending the PCI environment for embedded applications, CompactPCI has drawn from many of the strengths of the venerable VME bus environment — most particularly, the passive backplane architecture, the Eurocard form factor, and pin and socket connectors. Not long ago, some people were asking whether CompactPCI was even necessary.

There are still some differences between the two environments. VME, for example, supports up to 21 slots while CompactPCI is, for the moment, limited to eight. But in the main, the electrical and logical differences serve to underscore the differences in the markets the two bus technologies address. VME is a well-proven safe choice, designed for real-time applications and showing continuous performance improvements. CompactPCI, on the other hand, brings into play all the time-to-market and ease-of-deployment advantages of the standard PCI architecture, combined with a lower cost structure.

VME bus technology implements an entirely different electrical and logical layer originally designed as a non-multiplexed asynchronous bus. CompactPCI is also processor independent, but can take advantage of a range of low-cost silicon, and enjoys the advantages of plug-andplay operating system and applications software. As an extension to PCI, leading high-performance PC bus architecture, CompactPCI provides direct access to PC components and peripherals, and allows solutions to be developed on PCs for deployment in reliable CompactPCIbased embedded applications.

VME technology has continued to advance, adding support for highspeed 64-bit address and data transactions. For example, hot swap capabilities for VME are also under consideration. Lack of hot swap functionality is something of an Achilles heel for CompactPCI at the moment, as it takes time to wind down a PC-based network computer. However, by the end of 1997, CompactPCI should have had in place a hot swap specification defining various levels of compliance including dynamic software reconfiguration, and providing a platform for high-availability applications. At the same time, the two technologies’ bus peak performance will also be equivalent.

In the telecom space, meanwhile, a CT specification has defined a method of laying up to 4,096 voice channels on top of the CompactPCI architecture, in essence creating a voice subbus. As they use the same Eurocard form factor, CompactPCI and VME can be easily mixed in the same sub-rack for certain applications if necessary. Bridging between the two is not particularly difficult. Due to its large installed base, many vendors, and a vigorous trade association (VITA), VME bus technology will play a large role in industrial computing for many more years.

WHY COMPACTPCI NOW?
Clearly, the two standards will coexist for different markets. Where time-tomarket counts, as it increasingly does in telecommunications, and where open PC standards are a significant advantage, CompactPCI has much to offer. Telcos, cable companies, call centers, and other telecommunications concerns are often sitting on equipment that has been in place for twenty years and is in need of upgrades. New opportunities, such as the cable-modem systems for residential Internet access, are appearing daily. Telecommunications demands reliable embedded applications, based on open standards to take advantage of COTS (Connection Transport Service). CompactPCI melds proven PCI bus technology with the durability and strength of VME in a package that balances the best of both worlds.

Wayne Fischer is director of strategic programs at Force Computers, a Solectron subsidiary. Force Computers is a leading worldwide manufacturer of embedded systems and computer boards, serving OEMs in the telecommunications, industrial control, and command and control markets. Force serves the world through its 16 offices, maintaining corporate headquarters in San Jose, California, and European headquarters in Munich, Germany. An ISO 9001 certified company, Force is committed to total quality. For more information, visit the company’s Web site at www.forcecomputers.com

CompactPCI brings the open desktop standards of PCI bus technology to embedded applications, along with the Eurocard form factor and pin and socket connectors required to implement a PCIbased system effectively in the telecommunications and industrial computing milieu. But systems integrators and designers of telecommunications systems still need computers designed to take full advantage of this powerful new architecture. It’s not typically something they have the time to build, nor do homegrown solutions generally help propagate new applications. As the computing industry has proven time and again, a steady stream of new applications is the life-blood of any architecture.

As a result, companies are not only turning to outside vendors for CompactPCI computing resources, they absolutely require a robust marketplace underpinned by technical design expertise and high-vol-ume production. And, as processorand operating systemindependence is a major advantage of the CompactPCI architecture, it is imperative that systems be available based on the PC industry’s leading processor families.

Fortunately, this marketplace and this kind of choice is now taking shape. For example, the computers selected by Broadcom as the cornerstone for their CompactPCIbased pilot reference platform are embedded platform-specific systems from Force Computers, who also happens to be the second largest supplier of VME-based computing solutions. A subsidiary of Solectron Corporation, Force Computers is the first company to announce UltraSPARC IIi, Pentium, and PowerPC processor platforms on CompactPCI, with the UNIX-based SPARC implementations being of particular interest to telecommunications companies.