[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
[rtl] Re: Interrupt Latency..
>Interrupt latency measurements like this just don't mean much for
>systems in real applications.
I can't speak for other applications, but for mine it's *very* important.
My application can't stand, for example, the latency present in
M$-Windoze which I've watched range from 50us to tens of milliseconds.
While it's not "The mother of all measurements" it is a very important
onein many applications.
> Most real-time kernels can do about
>as well with a fast CPU.
That's probably very true, but can most real-time kernels provide the
wide range of development tools, wealth of application software,
and full source code like rtlinux and linux? The answer is simply no.
> If interrupt latency is your only requirement
>you should not use an OS.
I can't imagine any system where interrupt latency is the *only*
requirement. I also can't imagine any situation where I'd decide to
write the rest of the system software because interrupt latency is my only
real concern. To me, that would be a very silly reason indeed not to use
a system.
> You can just code such applications
>as stand-alone programs. This is done for many simple embedded
>applications.
While interrupt latency is important, so are developing programs, storing
data, generating displays, etc. etc. In my opinion, Linux/RTlinux is a
killer combination. I'd hate to have super fast intr. response but have
to write disk drivers, keyboard entry, display, etc. etc.
In my project I actually do have several "stand-a-lone" embedded
applications which do run without an OS. They run on PIC
micro-controllers. They can provide event timing control to 200ns, and
the newer 17c4x family to 120ns.
When you can't do it from rtlinux, just use a slave PIC 16c65a. They
nicely interface directly to the ISA bus on your computer with little
effort.
Thanks to all on this list!
C. Wayne Wright
--- [rtl] ---
For more information on Real-Time Linux see:
http://rtlinux.cs.nmt.edu/