In a message on 09-10-94, Mike Nice said to Peter Hansen:

MN>  OS/2 works fine for near real time control.  
  :
I know that much, but real-time is a continuum, not a binary choice. 
That is to say it stretches from very "soft", non-critical, low
performance applications, to "hard", critical, super fast stuff.  I know
OS/2 can cover the low end well, but my applications lie further along
the axis and I need to determine just how far you can push OS/2 down
that same path.

MN> I have an application that is text mode only and does some machine
MN> control-loop stuff. However, the control loop is protected by an
MN> external watchdog circuit because OS/2 has ...
  :
The fact that you can "protect" your system with a watchdog tells me it
is somewhat "soft".  That runs contrary to some people's definitions and
ways of thinking, but in my case if I miss one or maybe two iterations
of the inner loop (the 1000-2000 Hz one I mentioned elsewhere) I may
destroy a $3000 test specimen and probably invalidate all the test
results.  A watchdog would be a nice toy in my case but wouldn't really
add anything useful.

MN>with thread scheduling.  The PM Kernel is non-preemptible between the
MN>WinBeginPaint and WinEndPaint calls.  Some PM programs do all drawing
MN>during those calls, and even the highest priority time critical
MN>threads will not be scheduled.
  :
This is very interesting and I'm glad you mentioned it.  Yet another
danger of the current implementation.  I've already concluded I have to
use interrupts to implement the inner loop and much of the logic that is
currently handled in high priority tasks, but now I know I would have to
move even _more_ stuff to the interrupts... this is likely to become
untenable.

I have a feeling my particular plans will have to be changed slightly. 
It may be better if I use the current version of OS/2 to learn more
about the _other_ features (SOM, vmem, PM, etc.) and leave porting my
apps in earnest until the Workplace OS is available (on Intel chips,
too, I hope).  WPOS will avoid the problem you mention above and all
others like it, and would improve the interrupt latency to such an
extent that I will certainly attain my 2000 Hz without problems.

Thanks for the note, Mike.
Peter Hansen  ***  Engenuity Corporation  ***  Guelph, Ontario, Canada
Internet: peter.hansen{at}canrem.com    RelayNet:->CRS    FIDO:(1:229/15)
___
 * MR/2 * ***  Learn Esperanto, the international language.  ***

--- QScan v1.12b / 01-0348