Embedded Systems East Conference (long)
Hi folks,
I've been a fly on the wall of this thread for a couple months now, and really appreciate the exceptional quality of the content. For a public forum this is the best I've seen; and Allen, your analysis has been great.
Went this past week to the exhibit floor of Embedded Systems East Conference as an investor in Wind River, to learn more about the industry and their place in it. I was encouraged by what I saw, generally confirming the good impression of WIND I've gotten from this thread and other sources.
Here are some specific gleanings from the conference (note that this is not comprehensive - I may have missed some important stuff!). Feel free to clarify or question:
COMMERCIAL EMBEDDED SYSTEMS VS AERO/MILITARY: According to a guy from Rational Software, there are two separate industry sub-sectors related to embedded systems, one addressing commercial embedded systems, as represented at this conference, and another - with their own conferences and presumbly their own vendor pecking order - addressing aerospace and military applications of embedded systems. Within the aero/military realm, Ada is the dominant language, with C/C++ making inroads, notably with VxWorks from WIND and (maybe less visibly) pSOS from INTS. Comments below refer to what I learned of the commercial embedded realm.
RTOS DIFFERENTIATION: I heard that there are about 65 existing RTOS's, though many of these may be proprietary and narrow niche. At least a dozen were represented at the show. The most talked about were VxWorks and INTS' pSOS. Also visible were MWAR's OS9, Phar Lap's Realtime ETS Kernel, and various realtime extensions to other OS's. Three differentiators among RTOS's are (1) hardware support, (2) "hard" vs "soft" realtime, and (3) available development tools; these three are addressed below. (Other differentiators, not addressed here, include performance, multitasking, memory protection, scalability, standards support..)
1. Hardware support - what chip(s) does it run on? WIND's VxWorks is offered on all of the significant competitive hardware platforms, including PowerPC, 68K, CPU32, ColdFire, C166, MIPS, i960, x86, ARM, and SH-3. The only CPU I heard requested and apparently not yet support is the new ARM chip (StrongARM?), which is a superset of a previously supported ARM chip, and therefore likely (I'm guessing) to be supported soon. pSOS support is similarly broad, I believe; I don't know about OS9 or others.
2. "Hard" vs "soft" realtime support - does it provide deterministic realtime performance? A few OS's competing in the embedded market do not offer deterministic realtime performance, notably Windows CE; however, all direct competitors to VxWorks (such as pSOS, OS9) are "hard" realtime, as are the realtime extensions to Windows NT.
3. Development tools - how efficient is programming in this OS, using vendor and/or 3rd party tools? Overall, regard for WIND's Tornado tool set seems very high. The WIND field applications engineers seemed to run into some embarrassing glitches demoing in Windows NT, but these seemed more showmanship issues than buggy or poorly designed software. One comment, that Tornado "is certainly no Microsoft Developer's Workshop" but nonetheless the most full-featured among toolsets for embedded development, suggests to me that the whole sector still has a way to go in providing friendlier tools. WIND folks were griping (therefore may be scared) about INTS's pRISM tool suite, but they commented that pRISM is "1 « years behind" and that though announced last fall, pRISM has only just recently shipped a beta version.
WINDOWS SUPPORT: There's a crowd of products addressing Windows for the embedded and realtime markets, in every flavor imaginable. Here are some I heard of:
1. WIND has two flavors of Windows, WillowsRT for Tornado and Virtual-Time Software's RT-Win. WillowsRT for Tornado (they also sell it for some other environments) is distributed by WIND. (BTW, reportedly Willows was acquired by Award a couple weeks ago.) WillowsRT offers Windows emulation through its Willows Binary Interface (they demonstrated "Minesweeper" running in VxWorks) as well as native ("better than Windows") performance for Windows source code re-targeted to VxWorks via the Willows TWIN 32 and TWIN 16 libraries. These two routes are intended to "let developers combine source code and 16-bit binary objects, i.e. DLLs, VBXs, drivers and custom controls to achieve native performance without the need for access to vendor source code."
2. Virtual-Time Software sells RT-Win, which allows Windows 95 (not NT) to run as a task in VxWorks. This means, as I understand it, that a target embedded device could conceivably be self-hosting, running Tornado for development in Windows 95, while developing its target app. (Not clear to me that this is a big win.)
3. Imagination Systems (a software subsidiary of Nematron, industrial PC hardware vendors), offers Hyperkernel, a set of realtime extensions to Windows NT. Note that this is not intended for embedded systems, but for desktop or factory floor PC applications.
4. Phar Lap's tool suite supports download of Windows NT to the target system, using their RTOS, the ETS Realtime Kernel (this approach seems similar to Virtual-Time's downloadable Win95).
5. Microsoft Windows CE is embeddable though not "hard" realtime. However, MSFT shared their booth with others offering realtime extensions, including Annasoft and VenturCom. Annasoft offers, within their we-do-everything brochure, only RTKernel, "a powerful realtime multitasking system for MS-DOS," though they imply availability of realtime extensions for Windows NT. They push Windows CE, but don't mention any realtime support for it. VenturCom has RTX 4.1, a Windows NT extension providing realtime control, plus Component Integrator, a development toolset supporting Windows NT in embedded applications. The VenturCom stuff sounds interesting.
6. RadiSys offers Intime, supporting embedded Windows NT with realtime extensions based on their iRMX technology. They don't seem like a direct competitor to WIND, however, because they seem to primarily sell hardware - software packages, rather than software solutions offered across a broad range of platforms.
EMBEDDED INTERNET: Another hot issue (of course!) was the various means for thrusting the Internet into embedded devices, especially for the purpose of updating the device software. I didn't really dig into this issue at the show, but received a demo of WIND's implementation of Java in VxWorks. Looked cool.
CONCLUSIONS: Wind River Systems seems to be a well-diversified company within the commercial arm of the embedded systems market, with some visibility in the aero/military arm as well. WIND has the most talked about RTOS, VxWorks, and has customers with applications across the broadest spectrum of the market, from manufacturing apps, to set-top boxes, to NC's. The company has 70 "Tornado partners" (many displaying the "Tornado" statue at their booths) with hardware and/or software compatible with WIND's offerings. WIND's tools are the most highly regarded in the sector, though possibly lacking the friendliness of tools found in the PC software market (this may or may not be an issue with embedded engineers). INTS's pSOS is likely the most significant competitor, with reportedly a diminishing lead in marketshare; however, with an eventual release version of pRISM, INTS may become a more significant threat. Lucent's Inferno, a recent concern of this thread, had no visibility at the show. MWAR's OS9, another competing RTOS, is reportedly strong in the set-top box realm (due to their in-beddedness with Philips?), but does not have much visibility elsewhere. Realtime and embedded Windows are hot issues, and WIND seems strong here, though I'd keep an eye on MSFT and VenturCom. Embedded Internet is also an issue, and again WIND seems in good shape competitively.
FINALLY: Allen, I met David Larrimore (Wind VP-Marketing) and he had high praise for the analysis you've offered via SI. He tried to describe you to me on the chance that I might find you at the show, but "a fit, white, greying man in early fifties" was not enough to go on in that crowd. Maybe some other time.
David Schoenbach |
