NTP jumps time

[Steve Dubinskiy_1]

Need to understand whats going on. please help.
i have multiple servers that sync to the order of NTP servers. Others are ok except one that ALMOST every day jumps time:

Mar 6 08:57:53 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=3
Mar 6 09:01:14 servername xntpd[4055]: synchronized to 10.10.10.100, stratum=3
Mar 6 09:00:10 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=3
Mar 6 09:58:41 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=2
Mar 6 10:06:09 servername xntpd[4055]: synchronized to 10.10.10.100, stratum=3

Mar 8 11:39:10 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=2
Mar 8 11:48:46 servername xntpd[4055]: synchronized to 10.10.10.100, stratum=3
Mar 8 11:45:34 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=2
Mar 8 11:57:18 servername xntpd[4055]: synchronized to 10.10.10.10, stratum=2
Mar 8 11:59:28 servername xntpd[4055]: synchronized to 10.10.10.100, stratum=2

any ideas ?

[Ralf Seefeldt]

Hello Steve,

how is the time evolving over the day, is this server's clock slower and faster than the other servers? Is there a difference of 1 second even after 1 or 2 minutes?

May be, the local clock simply goes a littel wrong and the server. I think, I have sometimes seen some BIOS-Setting on a HP-PA-RISC server, to set the clock faster or slower.

The battery buffering the clock should not accect the clockspeed, even it is emty.

what are the settings for the ntp-clint?

Bye
Ralf

[A. Clay Stephenson]

Execute this command:
ntpq -p

The reach value should always be (except for the xntpd daemom is first starting) 377. If not then you are having communication problems with some of your NTP servers.

You should also examine the dispersion values which should be quite small. I also suggest that you configure a minimum of 3 timeservers so that NTP.

Read /usr/share/doc/NTP_Primter.txt for instructions on how to choose timeservers and interpret the ntpq output.

[Bill Hassell]

I hope you did not follow the typical Linux (and other Unix-ware) setup where ntpdate is run every few hours. When setup correctly, HP-UX uses ntpdate only once during bootup and xntpd adjusts the time in fractions of one second. Look in crontab for ntpdate and yank it out. Also check that no one has created some kludge script that tries to 'adjust' the time.

NTP (using xntpd) is extremely accurate and will never jump the time. If you stop and restart the NTP service using the startup script, then the time will jump based on the server you have coded into /etc/rc.config.d/netdaemons. You ntp.conf file should have 3 or more servers listed and one of them should be the server listed in the startup script. Bigger data centers have a firewall acting as the compnay-wide NTP server but if you can access the Internet with your server, just use the NTP pool. See:

http://www.pool.ntp.org/

[Steve Dubinskiy_1]

I also got a call with HP open for this.. so far no clues as to what may make it jump that wait it does now.

HP rep and I verified that network is good.
in my /etc/ntp.conf I have 2 entries for time server to sync against.

I verified cron and made sure no weird script is trying to manually adjust time.. xntpd daemon is running and as u can see from the copy of those log messages it is the one who syncs time incorrectly and then puts it back.

I appreciate all of ur ideas and points.. keep 'em coming, cause I'm sure one of them will be the answer i'm looking for.

only I was given is to restart xntpd daemon and see if the problem comes back

thanks,
Steve

[rick jones]

The folks in the usenet ntp group seem to suggest that two is really one of the worst number of servers to configure in NTP - it makes it difficult for NTP to detect which one is wrong if one of them happens to be. They seem to suggest at least sources of time.