Only a few years ago, the embedded arena was a quiet community where RTOS vendors vied against each other and against the predilection of potential customers to ignore them entirely by rolling their own software. But everything changed earlier this year when Microsoft disclosed its intention to turn CE from a downsized desktop clone into a battle-hardened OS with serious real-time capabilities.

Indeed, the Microsoft booth will be far and away the largest at this week's conference — a 50- by 100-foot spread that will devote half its ample territory to products from the company's burgeoning roster of CE software partners.

Marketing aside, Microsoft will use the conference to hold, under tight nondisclosure restrictions, its first broad software design review of "hard" CE, which is officially called 3.0. The company's customers and partners will be privy to the proceedings.

Development efforts are moving ahead on track, according to Tony Barbagallo, Microsoft's product manager for Windows CE. He said Microsoft expects to issue a "preview release" of CE 3.0 in the spring, with actual shipment next fall.

That could enable CE to gain credibility outside of the handheld PC, which was its original breeding ground. Palmtop computers such as Casio's Cassiopeia have undersold expectations. As a result, Microsoft is looking to new horizons in embedded apps, such as set-top boxes, factory-floor automation systems and precision robotics, to build a bigger user base.

"The best fit for CE today is in places where there's a visual user interface," said Paul Zorfass, lead embedded analyst at First Technology Inc. He cited consumer-electronics applications, along with smart phones, vehicle navigation systems and designs requiring interapplication connectivity.

Paul Rosenfeld, director of marketing at real-time embedded software vendor Microtec, believes the "bottom line" is that "there are some apps that CE does really well now; there are some that it might do better in a year, when CE 3.0 comes along; and there are some that it'll never do well."

CE 3.0 will be best-positioned for broader design wins if it can meet specific technical targets, analyst Zorfass said. "Microsoft has to do two things: CE 3.0 has to have a good level of determinism, and it has to be configurable." The latter would mean a truly modular implementation, probably including source code, which would enable customers to pick and choose the pieces of the OS they want to use.

Despite the paucity of CE-based deeply embedded apps so far, many industry players believe it's a mistake to dismiss the OS, because Microsoft and its partners are still priming the pump.

Indeed, to get momentum behind CE, Microsoft has strung together an unusual series of partnerships and marketing arrangements in the RTOS community. For example, Integrated Systems is offering support services for CE. Yet the company competes with CE via its own pSOS RTOS.

Most vendors explain such fence-straddling approaches by stating their belief that CE won't encroach on their bread-and-butter, deeply embedded applications. "CE 3.0 will be harder, as opposed to hard, so it addresses a bigger piece of the market," said Microtec's Rosenfeld. "But there's still a big difference between where [Microtec's] VRTX is targeted and where CE is aimed."

At the Embedded Systems Conference, however, CE will get a push that could prod it into one such arena: complex, ASIC-based embedded designs.

In that regard, Microtec will announce a deal with ARM under which Microtec will provide an integrated tool kit to enable deployment of CE on custom ARM-based embedded applications.

More important, the tools are intended to work in a hardware/software co-development environment. That's the kind of setup engineers are likely to use with ARM, argues Microtec.

"There are some unique challenges, such as debugging, that engineers face in deploying CE on custom ASIC hardware," said Rosenfeld. "In the traditional X86 world, you have standard boards. However, in the ARM environment, you don't even have standard processors; you have core-based ASICs."

Indeed, the use of application-specific processors is becoming the rule rather than the exception in the embedded world. Along with ARM, rivals include core-based offerings built around the MIPS Technologies architecture. Also in the picture are processors powered by Hitachi's SH3 and SH4 designs, along with PowerPC and X86 implementations. On the software front, all are supported by CE.

Applied Microsystems Inc. (Redmond, Wash.), for one, is addressing deeply embedded designs based on PowerPC, X86 and SH with its new Foundation-CE development suite. The company will highlight the suite at this week's conference.

Indeed, with so many platforms in play, the industry is realizing that serious embedded developers working in joint software/hardware environments won't be able to get by with old-style software tools. "There's multiple layers here," Rosenfeld said. "If all you're going to do is build apps, you can get away with a $500 tool. However, if you're building your own board, then you've got a problem, because those tools don't have the foundation for hardware debug.

"Looking further ahead, if you're going to do your own ASIC, you have another level of challenge, because you need $50,000 ASIC design tools."

While support for CE on embedded CPUs is important if the OS is to gain ground, there's another key element in the equation. Specifically, Microsoft must improve the real-time performance of CE 3.0. Yet even Microsoft acknowledges that CE may never deliver the guaranteed, ultrafast interrupt response times — features known as determinism and low latency — that are the equal of the best of the traditional microkernels.

That is one reason Microsoft isn't pushing CE for the toughest of the hard real-time apps but is pulling back half a step to tout the OS for many — but not all — applications. At the same time, Microsoft has expanded the discussion that's long been used to quantify interrupt response.

Indeed, Microsoft's Barbagallo argues that two numbers must be taken into consideration: "There's the time it takes when an interrupt comes in. Then there's the time it takes to respond to that interrupt."

The time it takes for CE to acknowledge that an interrupt has come in is under 10 microseconds, Barbagallo said. "On that playing field, we're exactly the same as all the other RTOSes. Where we in fact are slower is in the time it takes to register that interrupt if you want the OS to do something with it. In that scenario, we're under 50 microseconds. "In contrast, an RTOS like QNX or pSOS probably takes on the order of 10 to 15 microseconds to flip it over to the OS."

Still, Microsoft is giving no quarter to its competitors, noting that CE can use hardware-trapped interrupts for specific cases in which immediate service is mandatory.

RTOS counterpunch
Not surprisingly, Microsoft's rivals don't see things that way.

With CE 3.0, Microsoft will be competing most aggressively against two specific RTOSes: VxWorks, from Wind River Systems, and QNX, from QNX Software Systems Ltd. For their part, both vendors will provide their own respective counterpunches at this week's confab.

Wind River will come to the conference to showcase a major new release of its real-time development environment. Dubbed Tornado II, the integrated development environment comes with a new debugger and with integrated simulation capabilities to support hardware/software co-development. It's also closely coupled to the company's VxWorks real-time operating system.

"Tornado II provides a complete framework for our new networking and graphics environments," said Wind River president and chief executive officer Ron Abelmann.

To meet developers' needs, vendors are pursuing a couple of approaches. One tack assumes that embedded developers are better off with a fixed — and often heftier — set of GUI resources. The other approach gives developers the option of mixing and matching GUI components, the better to meet the parameters of resource-constrained designs.

For example, one of the raps against CE is that it is too bulky because it includes a full GUI. Microsoft disputes that assertion, claiming CE is completely customizable.

"If you want to build a full-up user interface that looks like Windows, then that takes up a fair bit of code, whether you build it with CE's primitives or with Zinc primitives," said Microsoft's Barbagallo. "There's no getting around that. However, if you want a smaller, specialized interface — maybe your application uses an LCD display — then you can do it with a very small amount of CE code.

"There's fundamentally no difference there between us and our competitors; it's just software functionality," Barbagallo added.

To prove its flexibility, Microsoft will showcase at the embedded conference a Unisys bank-check scanner based on CE that uses a downsized user interface.

For its part, QNX Software Systems is looking to blunt CE by playing its Java card. At this week's show, QNX will demonstrate a real-time Java application for robotics control running under its Neutrino RTOS.

For its part, Sun Microsystems' EmbeddedJava spec — which has some nine licensees — is continuing its movement from paper spec into code. At the same time, HP is proceeding with the incursion into the embedded Java world that it launched this past spring.

Neither Sun's nor HP's Java implementations run by themselves but must be paired with a traditional RTOS. HP's Java implementation has gained ground through license agreements with Integrated Systems, Lynx Real-Time Systems Inc., Microware and QNX. In addition, Wind River has endorsed — though it has not formally licensed — the HP technology. (Microware, QNX and Wind River are also licensees of Sun's EmbeddedJava.)

Microsoft, too, is in the Java picture. It has licensed HP's Java Virtual Machine and plans to integrate it into Windows CE.