Valdis.Kletnieks@VTTCF.CC.VT.EDU (Valdis Kletnieks) (06/11/91)
All you chronon counters with blue skins take note: AIX/370 1.2.1 - at the '0300' level (PTF U401375), IBM released APAR IX18142 which makes the 'adjtime()' syscall behave itself reasonably well. Our AIX/370 machine (a 3090-300) is currently running at stratum 3, and tracking our strat-2 boxen with about 2-3 ms offsets, and dispersions close to zero. The behavior of "slam the clock backwards several hours" is gone... The latest value in /etc/ntp.drift is -0.0098724403, which seems to indicate a reasonably stable clock... I think this is one of the longest-running problems I've ever had with a vendor. I reported this problem to IBM back on Sept 15, 1989.... Valdis Kletnieks Computer Systems Engineer Virginia Tech
jch@MITCHELL.CIT.CORNELL.EDU (Jeffrey C Honig) (06/11/91)
Is that ntp or xntp? Jeff
Valdis.Kletnieks@VTTCF.CC.VT.EDU (06/12/91)
> Is that ntp or xntp?
Jeff (and all):
It's "ntp" - I looked at 'xntp' once, had problems with undefined
kernel values, and bagged it. Maybe someday I'll get a reason
to go back and do another pass at xntp...
Anyhow, there's still a bug left in there someplace - on occasion,
somebody "forgets" and no adjtime() are done - I haven't traced out
whether ntp stops DOING it or if the kernel gets anti-social.
I usually notice it when the 'offset' gets up to 500ms or so, and
a 'kill -9 /etc/ntpd' and restarting it fixes it.
This behavior has been there for quite some time - I just haven't
had the motivation to chase it down yet...
/Valdis