RE: Ntp && clockspeed
Ethan Erchinger <[email protected]> Fri, 21 Mar 2003 09:04:20 -0800
| Newsgroups | gmane.comp.lib.libtai |
|---|---|
| Message-ID | <[email protected]> |
Hi all,
Using the clocksd stuff from below, if sntpclock fails to reach the ntp
server, sometimes it skews the time by many years. Check out this from the
current file, notice the large difference in the date:
@400000003e7b1f6f175c8994 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f7017646934 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f71176c5874 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f7217746ec4 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f73177ca454 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f7417849394 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f75178c8aa4 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f761794a0f4 sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f77179c941c sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f7817a4f88c sntpclock: warning: unable to read clock: timed
out
@400000003e7b1f7817a598b4 sntpclock: fatal: time uncertainty too large
@400000003e7b1f7817c6d09c before: 2003-03-21 06:19:26.398663000000000000
@400000003e7b1f7817c76cdc after: 1921-01-08 04:43:15.1263060690851990135
Give the following commands having been run, where clock_pick.sh runs
'sntpclock <ip>':
"
$CLOCKSPEED_BIN/clock_pick.sh \
| tee $CLOCKSPEED_HOME/adjust | $CLOCKSPEED_BIN/clockview
# show current "attoseconds":
$CLOCKSPEED_BIN/clockview < $CLOCKSPEED_HOME/etc/atto
"
Since I only see 1 set of 'before|after' marks it's clear that the output
was from the 'clockview < atto' line. So in theory no data was sent into
the adjust file. But then why would it skew?
Now, the reason it couldn't reach the server is my problem (ipchains). But
I could see a time when the internet connection is down and this would fail.
Any ideas on how to ensure this doesn't happen?
Has anyone else run into this problem?
Thanks,
Ethan
> -----Original Message-----
> From: John Peacock [mailto:[email protected]]
> Sent: Thursday, March 20, 2003 8:55 AM
> To: Ethan Erchinger
> Cc: '[email protected]'
> Subject: Re: Ntp && clockspeed
>
>
> Ethan Erchinger wrote:
> > I've seen recommendations of running clockspeed for time
> synchronization,
> > and running ntpd on a box or two to offer time services to
> a lan of hosts
> > that don't understand the taiclockd protocol
> (switches/routers/windows
> > boxes).
>
> Definitely check out these links:
>
http://fte.tegtmeyer.net/djbware/clockspeed_setup.html
http://www.guinix.com/software/clocksd.html
>
> Can anyone tell me if this is viable? The clocks of the servers I have
> running this way seem to be skewing abnormally. I have NTP looking at
only
> the local clock for time, without the 'prefer' keyword. So it shouldn't
be
> synching to that time.
This is exactly what we are doing. There will be a few problems (that I
consider minor) in that the NTP time will be (currently) 22 seconds off from
the
TAI time, as well as possible problems over leap seconds. I'm just going to
down the ntdp service prior to any leap seconds and bring it back up
afterwards.
If I can ever get the time, I want to write a taiclockd client for W2k and
Netware, so that I don't have to run NTP at all.
John
--
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4501 Forbes Boulevard
Suite H
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5748
(unnamed)
(application/ms-tnef, 3.7 KB) - not displayed