I wonder if engineers designing embedded computing systems are too concerned with fashion trends for their own good.
No, this has nothing to do with business-casual Fridays, but has everything to do with future generations of real-time, mission- and life-critical computers that by definition must be interrupt driven and highly deterministic.
At issue is VITA 46, an emerging technical standard that will define - through connectors, pinouts, and other details - a physical implementation of embedded high-speed serial switched-network fabrics for 6U and 3U single-board computers.
Plenty of engineers out there seem to think VITA 46 will render embedded parallel backplane databus architectures obsolete, and by extension ultimately will spell the end of VME, Compact PCI, and other parallel bus architectures.
I, on the other hand, see VITA 46 as a big Kenworth with The Future of VME and Embedded Computing painted on the side as it barrels down a nighttime highway. Many think they’re safely aboard, but I think they’re really just frozen in the headlights. We’ll know for sure probably within the next three to five years, when we see which embedded computing projects and companies succeed, and which ones get run over.
It could be true that VITA 46 is the key to moving embedded computing into the new age of switched-network fabrics. Maybe. It’s a big question, however, just which segments of embedded computing applications could, should, and will move over to switched networks.
Unfortunately too many program managers, marketing executives, and other technical experts are drinking the Kool-Aid. They conclude that switched-network fabrics and their VITA 46 implementations represent the only viable ways for embedded computing to move forward.
Some of these decision makers are under intense pressure to make commitments to VITA 46 and to switched fabrics - commitments that may prove to be difficult, if not impossible, to reverse without suffering financial and schedule damage.
Make no mistake; an abandonment of parallel databuses and wholesale adoption of switched fabrics would be a miscalculation that threatens program schedules, technology budgets, and perhaps even faith in the future viability of open-systems industry standards.
“The fabrics have been way over-hyped,” warns Ray Alderman, executive director of the VME International Trade Association - better known as VITA - in Scottsdale, Ariz. Switched-network fabrics, Alderman says, “are not a replacement for hard-real-time, tightly coupled backplane buses. No way these fabrics will ever replace them; you can’t get the determinism out of them.”
The contention of Alderman and other respected industry experts revolves around the technical strengths and weaknesses of switched-network fabrics, such as Rapid IO, Star Fabric, Infiniband, and Gigabit Ethernet.
These fabrics move data at blazingly fast speeds, but unlike parallel databuses like VME and PCI they are not predictable, or “deterministic,” down to the microsecond level. Switched fabrics “are not a panacea,” Alderman says. “They will not do hard real time.”
Switched-network fabrics, Alderman points out, also are far more software intensive than are parallel databuses. What this means is switched-network fabrics and VITA 46 ultimately will be well suited for large, clustered, multinode supercomputers, but not for real-time interrupt-driven systems such as fire control, self-protection systems, and command and control.
Systems designers need to understand the difference. If they don’t, switched fabrics and their VITA 46 implementation could be a train wreck going somewhere to happen, all because too many systems integrators are ready to bet the farm on a technology that is not best suited for all the applications at hand.
“Because of the over-hype of the fabrics, engineers will get deep into the fabrics and realize they have three years of software code to write and certify,” Alderman says. “They will back out. This will happen, and will be a great waste of money and human resources.”
Where switched-network fabrics will shine is in applications that require huge one-way data pipes, such as radar signal processing, electronic warfare, sonar-processing networks, and image analysis . . . and this is only WHEN the switched fabrics and VITA 46 come on the scene.
As of right there are no deployed implementations of VITA 46 - not even close.
“VITA 46 is still hammering out pin allocations, and is still not to the point of selecting the input power,” explains Douglas Patterson, director of business development at the SBS Technologies government group in Castaic, Calif., and who has long been active in VITA committees. The real applications of VITA 46, he says, are still three to five years away.
I see two weather systems converging here into what may become a perfect storm.
First, when all the technologies get sorted out - and when engineers come to realize that fast parallel databus architectures still have plenty of utility - switched fabrics and VITA 46 will be niche technologies, at best, with not a lot of critical mass in the market to drive them forward. That could further delay the first deployed implementations of these technologies.
Second, the longer that switched networks and VITA 46 implementations remain as laboratory prototypes, the more likely it is that smart designers in private industry will come up with something much better, easier, and less expensive than what we know of today as VITA 46.
If a better solution comes out of industry, at best it could be offered as a de facto industry standard, leaving a lot of wasted time and money in the wake of VITA 46. At worst, the better solution may remain proprietary, and demonstrate to other companies that proprietary solutions, not open-systems industry standards, are really the way to go.
There’s a lot at stake here. Program managers and systems integrators need to keep their eyes on the best technological solutions, not the most fashionable.
They need to remember that parallel backplane databuses still may be the best solutions for hard-real-time deterministic applications, while switched fabrics and VITA 46 may be the best solutions for data-intensive applications. What everyone needs to remember is that switched fabrics and VITA 46 do not represent a one-size-fits-all solution.
The future of open-systems embedded computing may be riding on it.
