[LinuxPPS] performance

gnu not unix gnu at wraith.sf.ca.us
Mon Oct 16 02:13:57 CEST 2006


Hi and hello--

In message <4532064F.3040800 at xs4all.nl> you write:
>Hello,

>As you can see at http://pindarots.xs4all.nl/mrtg/ntpstat.html the
>performance varies quite a lot.

Yes, interesting what you can see with plots. I plot alot...

>Especially in the month and year graphs you see a period of net-only,
>then nice GPS performance.
>Then (week37) stuff worsens, partially because of the previously
>reported link issue. After week 40 stuff improves but it does not get
>near the performance that I had 4+ weeks ago.

>ntpd was upgraded, every time just teh mainline of the program.
>Kernels were upgraded, same set of settings and patches.
>ntpd.conf did not change.


There is a recent post by Dave Mills that states the linux
kernel adjtime was tampered with, the hz was changed without
also making the associated loop parameter changes (my paraphrase,
given my understanding of ntp and  phase lock loops, beware :)


>Offset/jitter used to be .00x or maybe .0xx for the greater parts of the
>day, when there was reception. Now it is more like 0.xxx.

I have only a few days of running 2.6.18 linux kernel, but
my experience was that the system would take a long time to
close the offset, was very touchy, and could get into a state
where it never seemed to properly stabilize.

../Steven





More information about the LinuxPPS mailing list