October 23, 1997

The changing shape of PC-Based test and measurement

DAN STRASSBERG, SENIOR TECHNICAL EDITOR

As the industry prepares to bid farewell to the ISA bus, feuding form factors vie for the title of de facto PC-based-instrumentation standard.

ISA is dead; long live PCI! Too bad the transition to a new bus isn't that simple. One of the few things that suppliers of PC-based test-and-measurement (T&M) and data-acquisition hardware agree on is that, although the ISA bus isn't dead, it won't live forever. User demand may keep the bus alive for as long as 10 years--albeit with a steadily decreasing market share--but the handwriting is on the wall. ISA is obsolete, and, as time passes, the motivation for burying it will only grow stronger.

Threats to the bus's life come from several sources. You often hear of problems with making plug-and-play technology work with ISA-bus boards. Intel (Santa Clara, CA) and Microsoft's (Redmond, WA) PC 98 Design Guide (Reference 1) proposes the production as early as next year of desktop PCs that lack ISA slots. Modern PC main boards usually generate the ISA bus from the PCI bus. Suppliers of PCI-bus data-acquisition boards are quick to point out that most such ISA implementations, though satisfactory for low-speed devices, are too slow to meet ISA specifications. According to these companies, high-speed ISA-bus data-acquisition boards that work well in older PCs perform poorly in Pentium-based units.

Looking further ahead, the PC 98 Design Guide envisions desktop PCs as sealed boxes without features that resemble the peripheral-device slots of current desktop PCs. This vision may be practical for office PCs, but it can pose problems for developers of highly configurable PC-based modular-instrumentation hardware. In fact, the idea of sealed PCs almost immediately suggests plugging instrumentation modules into a chassis outside the PC. In other words, if no slots exist in the PC, the T&M hardware may have to provide them.

The USB (Reference 2) provides a way of attaching low-speed instrumentation peripherals to sealed PCs. In this case, low-speed means only as fast as 12 Mbps, about one-fifth the speed of the ISA bus. The IEEE-1394, or Firewire, high-speed serial bus performs a similar function for some higher speed instruments. But Firewire is less than 40% as fast as PCI: 400 Mbps (equivalent to 50 Mbytes/sec) vs 132 Mbytes/sec. Agreement is fairly widespread that, for a large number of applications, PCI is ISA's heir apparent. The real question is just which of PCI's several variants--that is, alternative form factors--will become standards in T&M and data acquisition?

Desktop PCI takes an early lead

So-called desktop PCI, the original form of PCI and the one whose size and shape most closely resemble ISA's, has jumped to an early lead. The majority of PCI-based data-acquisition-product introductions are in the form of desktop PCI. Suppliers of desktop-PCI-based T&M hardware point to the long and successful history of ISA-bus data-acquisition boards as evidence that the market doesn't need a new form factor.

These manufacturers assert that customers who need something more rugged than a desktop PC to house PCI-based instrument systems have the same option as ISA-bus-based systems have offered for years. Those customers can choose from dozens of industrial PCs. Such units now offer desktop-PCI slots. Moreover, besides instrumentation boards, the desktop-PCI slots can accommodate myriad low-cost boards designed for general-purpose PCs.

So, although desktop-PCI instrument suppliers concede that desktop PCI is sometimes less than ideal, they emphasize that you can compensate for a lot of deficiencies with the money the bus saves (see box "Economies of scale: hitching a ride on the PC bandwagon"). In addition, no other approach is nearly as versatile.

Moreover, only a few vendors appear to fully understand the diversity of customer needs. Because companies perceive those needs differently, several classes of PCI-based products are emerging. Some companies' products use technology, which, though physically different from desktop PCI, conforms to the PCI-bus electrical specification. These PCI variants include CompactPCI, a Eurocard-based PCI superset; Industrial PCI (IPCI), a European standard that shares CompactPCI's Eurocard format but differs from it in other ways; PCI mezzanine card (PMC), a mezzanine-bus version of PCI; and PC-104+, a stacked-card approach that targets small, fixed-configuration embedded system (Reference 3).

Industry groups and supplier consortia form another contingent that supports different approaches. Among these organizations is the PCI Industrial Computer Manufacturers' Group (PICMG), which backs CompactPCI and passive-backplane PCI (PBPCI). Some other industry groups promote IndustryPack, a mezzanine-bus standard; VXI, the 10-year-old, high-performance instrumentation bus that evolved from VME; and other buses that aren't based on PCI. The box "For more information..." provides contact information not just for vendors but also for industry groups.

And, of course, quite a few proprietary buses, particularly mezzanine buses, exist, some of which are optimized to work well on boards that plug into PCI. In some cases, the vendors publish the bus specifications for distribution to all interested parties--even other board manufacturers. The vendors behind these specifications would like nothing better than to see their buses accepted as broadly supported open standards. Indeed, several popular buses got started in just this way. Among them are CompactPCI and IndustryPack.

Besides the higher bus-transfer rates, another appeal of PCI is that it makes implementing intelligence on the I/O board much easier than ISA does. The fastest PCI boards and even some slower ones can act as bus masters. This architecture allows PCI boards to transfer data directly to host-processor memory without a DMA controller. Bus-mastering PCI data-acquisition boards also can dispense with deep onboard memory (often several megabytes on fast ISA-bus boards), thus considerably cutting cost. Moreover, when the host CPU has sufficient memory, such boards can acquire much longer records without gaps in the data.

Some vendors, such as United Electronic Industries and Keithley Metrabyte, use DSPs to manage the board's housekeeping functions. Others, such as Bittware and Datel, use DSPs for more traditional functions. Microstar Laboratories, a company that does not make PCI boards, likes to point out that it has been supplying intelligent data-acquisition boards for years. According to Microstar, a smart enough board can endow an ISA-based system with many of the attributes of PCI. Still, the combination of a fast bus and an intelligent I/O system may offer even more advantages (see box "I₂O: Does it hold H₂O?").

The contender: CompactPCI

Of the PCI variants, one that is starting to attract considerable attention is CompactPCI. Mechanically, CompactPCI boards look quite different from desktop-PCI boards and a lot like VME boards. Both VME and CompactPCI use passive backplanes, in contrast with desktop PCI, which uses an active main board--except in the PBPCI variant. VME and CompactPCI both use a 160-mm-deep Eurocard format, and both allow 100-mm (3U) and 233.5-mm (6U) board heights. Both also use faceplate-mounted I/O connectors on the card edge opposite the bus interface. (This arrangement contrasts with desktop PCI and ISA, whose I/O connectors are usually on an edge perpendicular to the bus interface.)

But, despite their mechanical similarities, CompactPCI and VME conform to separate electrical specifications that have little in common. And, despite their mechanical differences, CompactPCI and PCI have important electrical similarities. CompactPCI is electrically a PCI superset that adheres to the PCI spec.

In one key mechanical area, CompactPCI significantly differs not only from desktop PCI, but also from VME. CompactPCI uses board and backplane connectors that do not mate with those of VME. As a result, a system can't intermix VME and CompactPCI boards, and you can't accidentally plug one type of board into a slot intended for the other. VME connectors are not hard metric, whereas CompactPCI's gas-tight, pin-and-socket connectors are.

CompactPCI's connectors offer low capacitance that lets a bus accept eight loads, one of which is usually the host computer. Desktop-PCI systems can accept no more than four (often only three) loads. PICMG expects quick resolution of problems related to a connector scheme for live insertion and removal of CompactPCI boards. Backplanes for live insertion also accept standard CompactPCI boards but require you to remove power before inserting or removing such boards.

So far, the primary applications for CompactPCI lie in telecommunications and data communications. But CompactPCI advocates foresee a bright future for the bus in many other areas, including instrumentation. One reason for the optimism is high performance at reasonable cost. CompactPCI can achieve this goal, advocates say, because the boards use the same bus-interface ICs as those on desktop-PCI boards that are produced by the millions. Video and network-interface controllers are examples. Still, CompactPCI I/O boards cost 15 to 50% more than their desktop-PCI equivalents. One reason is that CompactPCI's bus-interface connectors cost more.

Current PCI embodiments in all form factors support a burst-transfer rate of 132 Mbytes/sec. A combination of a 4-byte-wide (32-bit) bus and a 33-MHz clock speed enable this performance. Board vendors report continuous transfer rates well in excess of 100 Mbytes/sec in systems whose host CPU and I/O boards use interface ICs of recent vintage. The PCI community has promised that, within a year, a 64-bit-wide bus and a 66-MHz clock will quadruple PCI transfer rates. The 64-bit bus seems like a sure thing; doubling the clock speed may take more work, though.

The PCI bus owes its speed in part to its compactness. The bus cannot extend as much as a foot from the IC that generates it. That IC is usually part of a PCI chip set, which itself must be within inches of the CPU (Reference 4). To create a system that contains more than four loads (eight for CompactPCI), you need PCI bridge chips that, in effect, create a hierarchical arrangement of PCI buses. You may be able to think of this hierarchy as a single bus, but if nanosecond delays and latencies are important to your application, you must learn about the subtleties of your system's main board or passive backplane.

For CompactPCI, the need for a short bus poses cost problems that lend credence to desktop-PCI-T&M-product manufacturers' claims of superiority. Restrictions on bus length prevent passive-backplane PCI systems from using general-purpose PCs' low-cost main boards. As yet, there is no cost-effective and technically appealing way to connect a desktop-PCI main board to any sort of passive PCI backplane. This limitation holds for both Eurocard-based CompactPCI and desktop-PCI-based PBPCI.

Therefore, PCI-based passive-backplane systems must use embedded CPUs whose form is the same as that of the I/O boards. Such CPUs cost considerably more than mass-produced office-PC CPU boards that offer equivalent performance.

Last month, National Instruments, which also makes desktop PCI T&M boards, announced a series of 3U CompactPCI-based T&M products. The company calls the product line PCI extensions for instrumentation (PXI) (see box "VXI and CompactPCI"). The announcement included a 60-mm-wide, 3U CompactPCI CPU based on a 166-MHz Pentium MMX µP. The CPU has 16 Mbytes of RAM; a 2.2-Gbyte hard drive; a display adapter with 2 Mbytes of VRAM; and serial, parallel, and USB ports. At $2995, however, the PXI CPU costs at least $1000 more than an otherwise-comparable desktop unit.

A rugged card cage that houses the CPU and as many as seven PXI modules costs an additional $1995. This cage includes a high-velocity air-cooling system and a 300W power supply that is designed to take more of a beating than those in most desktop PCs. A desktop-PCI partisan might say that if you use a general-purpose CPU, you can save the price of this cage, because the general-purpose PC includes a housing. That statement is somewhat unfair, however. The PXI cage is comparable to the housings of industrial PCs that cost substantially more than office PCs.

By using multiplatform extensions for instrumentation (MXI), an open standard that National Instruments pioneered, you can control VXI instruments from a desktop PC instead of using a more expensive embedded PC. Although National Instruments offers PCI and CompactPCI adapters that generate MXI, the PCI world offers no equivalent cost-saving way to link PBPCI or CompactPCI card cages to desktop PCs. Your first thought might be to use the Firewire serial bus. But, at least in its current implementations, Firewire does not support PCI's top speeds. Even in higher speed versions that have not yet emerged from the lab, Firewire can't keep up with today's PCI--and PCI is destined to get faster.

Instead of espousing a PCI variant, such as CompactPCI, Gage Applied Sciences is sticking with desktop PCI. However, the company now offers its own industrial PCs. According to Gage president Muneeb Khalid, the company's technical-support people found that customers were having too many problems building desktop-PC-based T&M systems. So, the company decided to build PCs its customers would find easier to apply.

Though simple to understand, some of the customer problems required a major PC redesign to solve. For example, on many desktop-PC main boards, the CPU chip is at the far end of the space for long boards in the PCI slots. (The far end is the board end opposite the connectors.) However, because a cooling fan must mount atop the CPU chip, the PCI slots can't accept full-length boards.

Whenever possible, T&M-board manufacturers avoid this problem by designing shorter boards. But that approach can increase the boards' cost. The GagePC 580 solves this problem and others and allows front access to the PCI boards' I/O connectors. Prices begin at $6995 for a unit with 32 Mbytes of RAM and a 200-MHz CPU. The price in-cludes a 350W power supply, an integral 10-in.-diagonal monitor, and a 2.2-Gbyte hard drive.

Sometimes, customer problems result not from thoughtless PC-main-board design but from users' own thoughtlessness or lack of understanding. Alligator Technologies supplies switched-capacitor antialiasing filter boards for PC-based data-acquisition systems. The company reports that it has repeatedly had to respond to customer requests for PCI-based lowpass filter boards. Several such requests have come from users who ought to know better.

If ever there was a poor use for a high-performance bus such as PCI, lowpass filter boards would seem to be it. Yes, the filter boards require programming. And yes, it's convenient to be able to program the boards via a bus. But with PCI slots so scarce, especially in desktop-PCI-based systems, users should be thinking of other locations for system components that can communicate well via a low-speed bus. One place for the filters might be on a signal-termination panel mounted outside the PC and connected to the data-acquisition board by a cable. The USB might be a way to supply the low-speed information to program the filters.

The scarcity of slots on a PCI bus creates other problems as well. Manufacturers find that they must cram more channels onto PCI data-acquisition boards than they had to supply on ISA boards. A common number of channels on an ISA board is 16 single-ended--also usable as eight differential. A few ISA boards have double these channel counts.

Some PCI boards are doubling the channel counts yet again. However, such high numbers of channels can require more connections than a desktop-PCI board's "front" panel can accommodate. (The front panel is the metal plate that is usually actually at the back of the PC.) In ISA systems, the way around such problems was to run cables from connectors in the center of the board to an additional front panel mounted in an unused ISA slot. If ISA slots are near the PCI slots, this scheme still works. However, if the only location for the extra front panel is another PCI slot, the approach accomplishes little. Interesting times lie ahead as PC-based T&M moves from the well-understood ISA-bus world to new, more capable technologies.

References

1. PC 98 System Design Guide, Intel Corp, Santa Clara, CA, and Microsoft Corp, Redmond, WA, Aug 14, 1997, http://developer.intel.com/design/pc98/.
2. Wright, Maury, "USB and IEEE 1394: pretenders, contenders, or locks for ubiquitous desktop deployment?", EDN, April 25, 1996, pg 79.
3. Quinnell, Richard A, "The mighty morphin' PCI bus," EDN, April 25, 1996, pg 58.
4. Rowe, Martin, "PCI bus cards and PCs must be compatible," Test & Measurement World, October 1997, pg 42.
5. Quinnell, Richard A, "USB: a neat package with a few loose ends," EDN, Oct 24, 1996, pg 38.
6. Quinnell, Richard A, "Universal Serial Bus Q & A," EDN, Feb 3, 1997, pg 206.
7. Wright, Maury, "Firewire unleashes the power of digital video," EDN, July 3, 1997, pg 44.

Please see the "For More Information..." box for a list of manufacturers.

| EDN Access | Feedback | Table of Contents |