Re: [CAD_CAM_EDM_DRO] Re: Pulse Gen

Jeff Barlow wrote:

> Jon,
>
> I sent Mariss a message off list in which I tried to explain this same
> issue. I suspect he is a bit naive when it comes to PC interfacing. The
> sad truth is that the PC interrupt system is very flaky, even with
> Linux.

Well, in general, OS's do their own stuff first, then worry about
servicing I/O interrupts. DOS doesn't have much housekeeping
to do, so it is not too bad, and you can take over the interrupts,
anyway. About the only one that really keeps on going is the
interval timer. Linux, without the RT patch, is much the same.

Linux WITH the RT patch is essentially an embedded system,
dedicated to the RT process, which allows wasted (idle) time to
be used by Linux. Interrupt latency is better than 10 uS even on
fairly slow processors, MAX! If you take that latency into consideration,
it is a fully-fledged hard real time system. This places a hell of a lot
of constraints on what the RT process can do - it can't make any
system calls at all, it is running in real mode, has access to the
entire machine's memory, etc. But, the RT process has as much of
the machine as it needs, when it needs it. Period.

If it misses even one interrupt, or takes longer than about 750 uS
to get around to servicing it, I WILL get an emergency stop on my
servo system. And, that has never happened, in 2 years, except
when the GUI locked up and I had to reboot. That did cause the
estop. (This was an old X-windows problem, corrected in early
1999.) But, in that case, even though the GUI was dead, and most
probably Linux, too, the RT process was merrily chugging along,
doing its work without fail.

Jon