[ntp:questions] autokey, NIST leap-seconds, and TAI on Win32

Serge Bets serge.bets at NOSPAM.laposte.invalid
Tue Nov 15 21:36:52 UTC 2005


There seems to be a problem with NIST leap-seconds file and TAI time on
Win32 NTP client: After autokey transfer, values for file stamp and TAI
offset are miscalculated.

First a Linux stratum 1 server running some days old
ntp-dev-4.2.0b-20051110.tar.gz, configured with all IFF/GQ/MV autokey
capabilities, and loading ntpkey_leap -> leap-seconds.3331497600 file
last updated 28 July 2005 and already containing the next 31 December
leap second infos. "ntpq -c rv server" shows "flags=0x800e3,
leapsec=200507280000, tai=32". Fine so far.

Second a Win2000 machine running latest Meinberg NTP installer version
ntp-4.2.0b at 20051016-1.1417-o-win32-setup.exe as stratum 2 autokey
client, here with GQ identity (same symptoms with IFF). It does not have
a ntpkey_leap file locally. After autokey/GQ negociation, "ntpq -c
readvar client" shows "flags=0x80063, leapsec=200506280000, tai=33".
One month less, and one second more. Feature?

The client cryptostat shows correct fs timestamp during autokey nego:

| 53689 68514.661 update ts 3341070114
| 53689 68514.662 leap 96 ts 3341023856 fs 3331497600
| 53689 68519.672 update ts 3341070119

BTW while experimenting with autokey on Win32, I had very frequently
service hangs, NTP becoming unresponsive, unstoppable, and unkillable.
Never 5 minutes without such a hang. Then I commented out "filegen
sysstats file sysstats type day enable" from ntp.conf, and no more hangs
since then. Seems autokey *or* sysstats do work, but not both?

Serge point Bets arobase laposte point net

More information about the questions mailing list