God <root%bostonu.csnet@csnet-relay.arpa> (11/18/84)
Reply to Geoff Kuenning's comments: You misunderstood me. The hesitation was experienced only by a technician using a terminal to fill out medical questionnaires while another person was having a lung test NOT BY THE REAL TIME PROCESS!! Obviously someone has to suffer in favor of the real time demand. To be more specific, we were sampling a 14-bit A/D at about 1KHZ for about 5-10 seconds and performing averaging and volume to time-domain conversions within the driver at sample time. IE. the stuff had tighter restraints than you describe by at least a factor of 4. Well, a factor of 4 is splitting hairs, but it sounds like what we were doing is well within your definition of 'real-time'. The point is, either your processor is much much faster than your real-time needs or you have more than one processor or you allow delays in your non-real-time processes in favor of your real-time processes. My only suggestion was that given the price of small processors today (some of which are plenty fast enough for most real-time needs) the second choice might be worth considering. That, for example, is why DEC sells their Lab Peripheral System even tho VMS claims to be capable of real-time...why tie up a $500,000 VAX when a $15,000 system will do the real-time job? -Barry Shein (not Schein) Boston University
geoff@desint.UUCP (Geoff Kuenning) (11/22/84)
In article <5877@brl-tgr.ARPA> Barry Shein writes: >Reply to Geoff Kuenning's comments: > >You misunderstood me. The hesitation was experienced only >by a technician using a terminal to fill out medical >questionnaires while another person was having a lung test >NOT BY THE REAL TIME PROCESS!! Obviously someone has to >suffer in favor of the real time demand. Oh. Oops. >To be more specific, we were sampling a 14-bit A/D >at about 1KHZ for about 5-10 seconds... >it sounds like what we were doing is well within >your definition of 'real-time'. I would agree. However, let me point out that if you used a DMA A/D device, your actual time constraint was the frame time, not the sample time, since CPU tardiness only affects the DMA sampling rate if you miss switching buffers. If you did not use a DMA device, I wonder what sort of things you did to be sure that you were actually sampling at 1KHz reliably, with no missed ticks once a second as Unix spent 4 ms running a lot of clock- dependent stuff. (That 4 ms figure is from a measurement on a 68000 Unix). >The point is, either your processor is _m_u_c_h _m_u_c_h faster [emphasis added] >than your real-time needs or you have more than one processor >or you allow delays in your non-real-time processes in favor >of your real-time processes. My only suggestion was that given >the price of small processors today (some of which are plenty >fast enough for most real-time needs) the second choice >might be worth considering. That, for example, is why DEC >sells their Lab Peripheral System even tho VMS claims to >be capable of real-time...why tie up a $500,000 VAX when >a $15,000 system will do the real-time job? The last time I looked, the $15,000 LPS was 11-based. That means that DEC is selling it with either RT or RSX, both of which are excellent low-overhead real-time operating systems. Which is exactly the point, because if you have good software you don't need a processor that is "much much" faster than your real-time needs. To me, half the fun of a real-time project is figuring out how to get the processor you can afford to perform to your requirements (though I wouldn't want to go too far in that direction...). -- Geoff Kuenning First Systems Corporation ...!ihnp4!trwrb!desint!geoff
henry@utzoo.UUCP (Henry Spencer) (11/24/84)
> The last time I looked, the $15,000 LPS was 11-based. That means that DEC is > selling it with either RT or RSX, both of which are excellent low-overhead > real-time operating systems. ... Neither one, of course, is much good as a timesharing system. So comparing them with Unix is silly. Using one of them reduces to the standard solution for doing heavy real-time work on any time-sharing system: buy a satellite processor to do the hard stuff. Who cares what's running in the satellite? -- Henry Spencer @ U of Toronto Zoology {allegra,ihnp4,linus,decvax}!utzoo!henry