[LinuxPPS] LinuxPPS comparable to FreeBSD here

[Hal V. Engel]

On Friday 18 June 2010 01:23:22 am Miroslav Lichvar wrote:
> On Fri, Jun 18, 2010 at 12:25:24AM +0200, Remco dB wrote:
> > Compared to my (also thermostatised) FreeBSD system (with hardpps)
> > LinuxPPS performs almost equally.
> >
> > See http remco.org/ntp
> 
> Interesting, what polling intervals are used?
> 
> > So I wonder, would adding hardpps _improve_ the LinuxPPS performance,
> > considering 'general hardware boundary conditions'?
> 
> That depends on the ratio between the PPS noise and the system clock
> noise/wander. If you have a thermally stabilized crystal, switching to
> kernel consumer (which is basically a FLL running at poll 0) will
> probably not help. You might want to try different polling intervals
> though.
> 
> FLL should perform better than PLL when the PPS signal has very low
> jitter and/or the crystal temperature varies a lot (or it's a low
> quality crystal).
> 
> You might also want to try chrony which uses a linear regresion model.
> It usually performs better than both PLL and FLL.
> 

Looking at the chrony email lists I see that you have been working on it.  In 
particular I noticed that you created a set of temperature compensation 
patches.  Are these based on the work of Mark Martinec 
http://www.ijs.si/time/temp-compensation/ ?

Also I notice that refclock support was added to chrony earlier this year.  
This appears to be limited to three types of refclocks PPS, SHM and SOCK.  Are 
there plans to add more drivers?  For example, the NTP NMEA and Oncore drivers 
are probably the most widely used drivers by members of this list and it would 
be nice to have these devices supported natively.  I sometimes boot over to 
Windows on my machine and on Windows the ntp refclock support is limited so I 
have to use the PPS driver with my Oncore.  This does work but not nearly as 
well as the native driver (slow to fully sync up,  larger offsets and requires 
the use of external servers).

Hal