On 29/01/08 18:19, Gilles Chanteperdrix wrote:
> On Jan 29, 2008 9:09 AM, Juan Antonio Garcia Redondo
> <[EMAIL PROTECTED]> wrote:
> >
> > On 28/01/08 14:19, Gilles Chanteperdrix wrote:
> > > > >
> > > > >
> > > > > > > No mystery: hitting a key on a telnet session causes an interrupt
> > > > > > > masking section of 110us, you see it as the maximum if you never
> > > > > > > observed longer masking sections, but it is not the maximum if you
> > > > > > > observed longer masking sections.
> > > > > >
> > > > > > OK, but why the masking section on linux side affects to xenomai
> > > > > > side ?
> > > > > > Another thing I don't understand is why when the system has load
> > > > > > (above
> > > > > > I'm talking about calibrator but the same occurs with dd
> > > > > > if=/dev/zero
> > > > > > of=/dev/null), the effect seems to dissapear.
> > > > >
> > > > > It is probably not a masking section on linux side but rather a
> > > > > masking section on I-pipe side. Anyway, the effect does not disappear:
> > > > > it means that the cache effects cause larger latencies than the
> > > > > ethernet interrupt, but maybe I did not understand what you explained.
> > > > > The results you obtain with no load are simply irrelevant.
> > > >
> > > > I'll try to explain it better:
> > > >
> > > > o Without load I run ./latency -t0 -p500.
> > > > RTD| 33.182| 53.479| 67.976| 0| 31.250| 77.319
> > > > RTD| 43.170| 53.479| 67.654| 0| 31.250| 77.319
> > > > RTD| 41.881| 53.479| 67.332| 0| 31.250| 77.319
> > > > RTT| 00:02:07 (periodic user-mode task, 500 us period, priority 99)
> > > >
> > > > o Each time I press a key (over a telnet session) I can see the lat_max
> > > > field increase on 40 to 50 us aprox.
> > > > RTD| 33.505| 53.479| 71.842| 0| 26.739| 77.319
> > > > RTD| 40.592| 62.177| 123.067| 0| 26.739|
> > > > 123.067
> > > > -------
> > > > \_________: Key pressed
> > > > RTD| 50.579| 53.479| 73.775| 0| 26.739|
> > > > 123.067
> > >
> > > This is where you are wrong:
> > > - first, let me repeat it: test made without load are irrelevant;
> > I can't agree with you. When we stress a system with load is, as far as
> > I know, because usually, the large latencies don't appear on a quiet
> > system. Here we have a case where a large latency (the lat_worst number
> > I've gotten after more than 7 hours with the system fully loaded is even
> > less than this) appears on a quiet system and directly related to an
> > external event.
> >
> > > - second, an event has no relative effect on max latency, its effect
> > > is absolute: pressing a key over a telnet session causes, for unknown
> > > reason, a masking section of around 130us, which happens to also be
> > > the worst case latency that we measured properly with a loaded system.
> > I pointed out the lat_max field because, if you keep the latency test
> > running and hit a key on a telnet session, you can easily see how the
> > lat_max increase each time you hit the key, while the lat_worst increase
> > depends on the former history.
> >
> > >
> > > Now, if you want to know why you get such a masking section, you are
> > > free to investigate.
> >
> > I'll try to do it.
>
> Does the network driver use the AT91 PDC ? If yes, and if it is
> possible to disable it, could you try disabling it ?
Yes, the network driver uses DMA and is not easy to disable it. Anyway I
have, in my custom board, an additional ethernet chip (smc91x). I've
done the former tests with the smc91x (LAN91C111), which can't use DMA,
and the behaviour is similar.
Regards,
Juan Antonio
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help
