[LinuxPPS] Stats & architectures

[Robert Jenkins]

> From: giometti at enneenne.com Sent: 11 March 2008 12:53

> Subject: Re: Stats & architectures
> 
> On Tue, Mar 11, 2008 at 10:23:59AM -0000, Robert Jenkins wrote:
> > Hi Rodolfo & all,
> > 
> > I'm still having tremendous problems trying to get PPS to 
> work properly on
> > my new server..
> > 
> > Two questions:
> > In the present PPS implementation (V5.x), what should 
> ntptime show when PPS
> > is in lock? On my system it has never shown anything other 
> than 'status 0x1
> > (PLL)' where older PPS versions (on my previous system) 
> used to somewhat
> > different info?
> 
> Just the same as before I suppose...

Can you please copy the ntptime output from a system with PPS in lock so I
can see what it should look like?


> 
> > Is anyone else using PPS on an x86_64 dual-core system and 
> getting good
> > results?
> > 
> > ------
> > With an earlier kernel (2.6.23) I got good stability (both 
> error & jitter
> > settled at 0.001) but it never showed any status other than PLL.
> > 
> > I rebuilt everything using the 2.6.24 kernel and now it 
> jumps about at
> > random - it settles down, sometines to .001 for a while 
> then jumps, with
> > anything from a few uS to 3mS increase in error.
> > 
> > (This is both with & without the kernel patch. 
> > That patch does have a significant effect, the frequency is 
> around -95ppm
> > without the patch and -66ppm with it, but either way is 
> still jumps and
> > jitters.)
> 
> Are you using the same LinuxPPS version on both systems?

The 'old' system was using a ppskit-lite patch. That locked perfectly and I
remember the ntptime output changed noticably once it attained lock.
It was based on Centos 3 i386 on a P3-866. (The machine has died so I cannot
compare the displays).

The new machine is based on Centos 5 x86_64 with an Athlon BE2300
(approximately a low-power Athlon 64 X2 4000).
Everything I've tried for PPS on this is based on patches etc. from enneenne
wiki.

With the 2.6.23 kernel plus ntp-4.2.4p2 and seperately downloaded patches,
it got a stable lock after some hours but the status never changed to
indicate PPS was being used. Once it had settled I never saw it above 0.001
error, but the lack of PPS indication puzzled me.
(The PPS device is present & edges are being seen by the test client).


I rebuilt everything from scratch using the 2.6.24 stuff and now it appears
to jump just as it should get into lock.
I've also tried the git kernel, and ntp-4.2.4p4 with the p2 patch (which
applies cleanly) but the occasional jump / jitter is still there.

Robert.


> 
> Ciao,
> 
> Rodolfo
>