Re: [chrony-dev] PPSAPI: kernel consumer

[ Thread Index | Date Index | More chrony.tuxfamily.org/chrony-dev Archives ]


On Tue, 9 Feb 2010 10:31:28 +0100
Miroslav Lichvar <mlichvar@xxxxxxxxxx> wrote:

> On Tue, Feb 09, 2010 at 02:04:39AM +0300, Alexander Gordeev wrote:
> > I still don't get the philosophy/complexity problems. BTW I don't
> > have any code to add to chrony.
> > 
> > I hope other chrony developers will still consider adding kernel
> > consumer support a valuable feature or maybe just add it to allow
> > the user to choose the possibilities.
> 
> I wouldn't mind adding support for it if it was just setting a flag,
> like in ntpd which in normal operation doesn't read anything from
> kernel back. In chrony the frequency and phase adjustments would need
> to go through several layers in opposite direction, it doesn't seem to
> be easy to implement.
> 
> I'm still not sure what would be the benefit.

One step further to full ntpd replacement?
Sure, second long convergence is not a killer feature since chrony can
sync quite fast too. So if it's hard to implement than forget about it.
I just don't want to be obliged to use either ntpd + kernel discipline
or chrony. chrony + kernel consumer can be much better. Well, maybe
some day I'll do it myself.

> > > Running at a shorter poll interval may make chrony
> > > more responsive and not decrease the accuracy. Try it.
> 
> I wouldn't recommend using poll below 2. It's likely to be unstable
> due to the kernel interface deficiencies. The most extreme setting
> you might want to try is "poll 1 filter 1 dpoll -1".

OK, thanks. Surely I'll test chrony a lot in the future. :)

> BTW, I was thinking about the adjtimex problem with detecting when
> were the values actually updated and it might be possible to find the
> edge by setting adjtime to slew 1 us and then run readonly adjtime in
> a loop every millisecond or so until it reports that there is no
> offset left.
> 
> This could make the measurements done while slewing much more
> accurate. Not sure how reliable it would be, I'll run some tests.

Sounds like a huge hack, sorry...

-- 
  Alexander

Attachment: signature.asc
Description: PGP signature



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/