From: peter@engcorp.com (Peter Hansen) Reply-To: peter@engcorp.com (Peter Hansen) Newsgroups: comp.realtime Subject: Re: OS/2 as an RTOS? Date: 20 Nov 1995 04:38:24 GMT Organization: Engenuity Corporation Message-ID: <48p0o0$ncn@granite.sentex.net> References: <48g94l$ph@news.internetmci.com> In <48g94l$ph@news.internetmci.com>, "Paul H. Stephan" writes: >Does anyone have any experience or opinions regarding the use >of OS/2 (Probably Warp) as an RTOS? Is it deterministic? How >is the interrupt latency? OS/2 can be used for the softer varieties of real-time tasks. It is not strictly deterministic. According to the bible on the subject, "The Design of OS/2" by Deitel and Kogan (ISBN 0-201-54889-5), "the system guarantees that there is a maximum interrupt disable time of 400 microseconds, and that time-critical threads are dispatched within 6 milliseconds of becoming ready to run." On the other hand, another section in the same book claims the 6ms figure is actually 4ms.... Typical latency times are much less and depending on the configuration the dispatch delay can be higher or lower. If you have complete control over the system (e.g. no users firing up unexpected tasks) and enough memory to implement a swapless system, you can use OS/2 successfully for certain kinds of real-time work. By the way, the PowerPC implementation of Warp is microkernel-based and is therefore likely to have much lower interrupt latencies, greater reliability and stability (though Warp is already excellent), and may be more suitable for what you intend. IBM and DEC are even, according to reports from some months ago, working on an actual RT implementation but for all I know that's pure vapour. I have no experience with the PPC version yet. ------------------------------------------------------------------------ Peter HANSEN Engenuity Corporation peter@engcorp.com Guelph, Ontario, Canada http://www.sentex.net/~engcorp/peter/ ------------------------------------------------------------------------ From: bruce@faxmail.co.nz (Bruce Simpson) Reply-To: bruce@faxmail.co.nz Newsgroups: comp.realtime Subject: Re: OS/2 as an RTOS? Date: Fri, 17 Nov 1995 08:30:11 GMT Organization: FaxMail Technologies Message-ID: <48hdpb$7u8@news.express.co.nz> References: <48g94l$ph@news.internetmci.com> "Paul H. Stephan" wrote: >Does anyone have any experience or opinions regarding the use >of OS/2 (Probably Warp) as an RTOS? Is it deterministic? How >is the interrupt latency? I'd recommend against OS/2 for anything other than moderately "soft" real-time. There appear to be fairly significant portions of the kernel which are not pre-emptable and this affects scheduling latency quite significantly. You can circumvent this to a small degree (but at the expense of significant added complexity and limited functionality) by puting most of the code which must be highly deterministic) into a device driver. Of course that requires that you buy the DDK and a suitable compiler (unless you want to do it *all* in assembler :-() Many people (including myself) have produced real-time systems in OS/2 but I've got to say that practical experience proves that it's an awful lot easier and less frustrating to use the right tool for the job. I built one real-time system in both QNX 4.2 and OS/2. The OS/2 version took a lot more effort, required almost three times as much RAM and far less "headroom". The QNX version by comparison was a *breeze* to write and outperformed the OS/2 system in every way. OS/2 is a desktop OS designed for use by devices (people) with relatively high latencies :-) *----[Fixed-price software development over the net ]----* | bsimpson@iprolink.co.nz or bruce@faxmail.co.nz | *--[C/C++, Win, OS/2, POSIX, device-drivers, fax, comms]-*