RE: EMC for windows

["ERIC KELLER" <eekeller-at-psu.edu>]

John,
I would never say anything bad about Mach1, but it has been a mystery to me how
it can work as well as people say it does.  I think the combination of fast
computers, an open loop stable system, and incredibly good (unimaginably good
for windows) performance in his driver makes it work where many others have
failed.  There is a lot of industry experience that says that in the general
case, real time windows is an oxymoron.  Thats how people can charge $xK for
their real time addons.


The real time windows add-ons actually work the same way that RTLinux does, i.e.
they take over control of the interrupt controller and run windows as a task. 
Victor Yodaiken, the top dog in RTLinux has made noises about suing the vendors
of windows rt add-ons, but I think he just does it for fun or self-protection. 
And windows can actually take the real-time packages down, from what I have
heard.

I believe I understand what Art has done fairly well, because I was headed down
that path with my fatigue machine controller until I decided to get done with
my thesis and wrote the controller in DOS.  This was when RTLinux was pre .9. 
I never got far enough in my windows driver writing that I could play around
with the priority level of my driver, but pumping up the priority level of your
application does not protect you.  I have a fairly large collection of smashed
aluminum to prove that.  My winNT controller would work good for weeks at a
time, and then someone would open a browser window during a long boring test
and kapooie, smashed aluminum.  This would usually not happen, but enough to
make it worthwhile to seek alternatives.

Under stock windows, there is no such thing as a real time driver, but you can
increase your priority.  I'll still have to see other people run a closed loop
servo system on windows for quite a while before I would be convinced to change
my tune.

Eric



On Wed, 29 Jan 2003 16:25:59, jmkasunich-at-ra.rockwell.com wrote:

> 
> 
> 
> 
> "Erik Keller" wrote:
> 
> > There is one counter-example which shows that under certain
> > circumstances windows will work in real-time: Mach 1, which
> > is the twin of EMC hosted under windows.  My theory on why
> > that works is that, by definition, a step and direction system
> > is open loop stable.  Thus, when bill gates has internet
> > explorer dump your hard drive over the internet to Redmond,
> > nothing bad happens, because no step/dir pulses are output.
> 
> It's not that simple.  If the steppers were moving fast when
> windoze started talking to Redmond, the pulse train would
> stop abruptly.  It wouldn't send the table flying, but it would
> cause lost steps because the motors wouldn't stop instantly.
> 
> Since Mach1 manages to avoid lost steps, Art must be doing
> something more sophisticated.  I think he uses a real-time
> device driver to handle the low level stuff.  The driver runs
> at a higher priority than almost everything else in the
> computer, so even when windoze itself takes a nap, the driver
> keeps doing its thing.
> 
> In other words, it's not "impossible" to do CNC under windows,
> just difficult.  Unmodified windows is no better or worse than
> unmodified Linux for real-time stuff.  With Linux, the answer
> is the real-time extensions (RT-Linux or RTAI).  With windows,
> the answer is a real time low level driver.
> 
> John Kasunich
> 
> 
> 
> 
> 
> 
>