[ntp:questions] Re: radioclk DCF doubling

Serge Bets serge.bets at NOSPAM.laposte.invalid
Sun Dec 18 21:11:16 UTC 2005


 On Tuesday, July 19, 2005 at 19:58:42 +0200, Serge Bets wrote:

>> Now the idea: Patch radioclkd to feed 2 SHM slots with same values
>> from only 1 receiver
>| server	maxpoll 6	prefer
>| fudge	time1 0.020	refid DCF1
>| server	minpoll 10
>| fudge	time1 0.020	refid DCF2
>|      remote           refid      st t when poll reach   delay   offset  jitter
>| ==============================================================================
>| *SHM(0)          .DCF1.           0 l   41   64  377    0.000    0.116   0.238
>| +SHM(1)          .DCF2.           0 l   10 1024  377    0.000    0.116   0.194
> And sync survives well over an hour with radioclkd stopped. I seem to
> get the expected low offsets and high autonomy. Now I wait next
> thuderstorm and will update you.

Update after some monthes and several thunderstorms using this twin
radioclock. The prefered one at poll 6 gave me low offsets, and the
second at poll 10 really gave me extended sync during periods of bad
signal reception. Up to 6 hours of storm, up to 80 consecutive minutes
of lost signal, and no sync loss. Fine! But I stopped the experiment,
and consider it failed.

The problem is that during stormy days, with the random rate of
parasited undecodable minutes, ntpd clockhops several times between
SHM(0) and SHM(1). And these clockhops perturbate the frequency, giving
it a sometimes important kick. This in turn perturbates the offset for a
long time. Result: I get a degraded accuracy during several hours, even
long after the return of good DCF signal.

Since some weeks I experiment another setup, with one only SHM(0)
radioclock at poll 6, and the new orphan mode that appeared in recent
ntp-dev versions (now using 4.2.0b at 1.1430-o). The clockhoping between
SHM(0) an orphan mode seems to not have bad effects on frequency
stability. Only bad effects are transient wrong leap bits. So I wait
with confidence for the next thunderstorm, but hope it will not be the
31 around midnight. ;-)

Serge point Bets arobase laposte point net

More information about the questions mailing list