lkml.org 
[lkml]   [1999]   [Feb]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [offtopic] Re: 2.2.2: 2 thumbs up from lm
On Fri, 26 Feb 1999, Arvind Sankar wrote:

> On Fri, Feb 26, 1999 at 08:06:01AM -0500, Richard B. Johnson wrote:
> > On Fri, 26 Feb 1999, Neil Conway wrote:
> > [SNIPPED]
> > >
> > > But how can you promise the customer that your system will respond to
> > > events while they are current if you DON'T control the latency?
> >
> > I have a CPU which does nothing but poll a port. I want to detect
> > when the sun comes up which happens once a day. There is a photo
> > controller relay connected to that port. I need to know this time
> > within one second.
> >
> > Therefore I need to poll this port at least twice per second.
> > Twice comes from Shannon since this has become a sampled system.
>
> Talk about overkill... btw, I think you might be referring to Nyquist's thm?
> Or is it just called the Nyquist frequency, but the theorem was by Shannon?
>

It is overkill, but the specification was not that we need to know that
it occurred within some time, but the actual time within a second so
you need to check it twice.

And yes Shannon worked for Ma Bell and published a lot of stuff. Basic
"digital information theory" was first published by Shannon,
"Communication in the presence of noise", Proc. IRE, 37, 10-21, Jan 1949.
The bottom line is that information is conveyed by changes. To detect a
change, one must sample an event at least twice as fast as it is expected
to change.

This sample rate is called the Nyquist rate. Nyquist preceded Shannon,
but in his day there was nothing called "digital" (he didn't do
windows!!).

> >
> > Do we care how much time it takes to poll this port as long as
> > the specified conditions are met? This is the latency and, in
> > fact we don't care as long as the required specifications are
> > met. If the latency is 1/2 second, we can't guarantee that the
> > port is polled twice per second. However, if the latency is
> > one microfortnight less than 1/2 second we can.
>
> and this is exactly what is meant by controlling latency. I find it difficult
> to put any other interpretation on that phrase other than by seriously
> distorting the English language.
>

Yep, but the latency required is defined by the application, not by
maintaining some fixed sampling (context switching) rate.

> Your system in this case guarantees latency < 1/2 second. Hence it is a
> `realtime system', according to the most common definition, which is that a
> realtime system is one that gives guaranteed upper bounds on response time to
> whatever events are relevant to the system.
>
> A guarantee, is, of course, with some probability bounds, which should be part
> of the system requirement specifications.
>
Yes.

> An operating system on its own cannot be classed as realtime, without making
> assumptions about the hardware it is going to run on.
>

Yes exactly, and the application requirements, which was my original
premise. This is why a "Realtime Operating System" means nothing. It's
just a buzzword. It could mean that it contains something useful, but
may not.

It is only when one analyzes the application specific requirements that
one can make a determination about an Operating System's suitability
for a purpose.


> eg. if somebody has a PC with a suspend-to-disk mode, it would be difficult to
> guarantee anything about response time to signals on ports if he started
> hitting the suspend button.
>
> -- arvind
>

Cheers,
Dick Johnson
***** FILE SYSTEM WAS MODIFIED *****
Penguin : Linux version 2.2.1 on an i686 machine (400.59 BogoMips).
Warning : It's hard to remain at the trailing edge of technology.
Wisdom : It's not a Y2K problem. It's a Y2Day problem.



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:50    [W:0.065 / U:0.924 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site