[ntp:questions] Re: linux system clock egregiously slow!

David Wuertele dave-gnus at bfnet.com
Fri Jan 23 20:31:51 UTC 2004


DLM> Configure ntpd with a well behaved source. Put "disable ntp" in
DLM> the configuration file.

I'm guessing that a stratum 1 NTP server counts as a "well behaved
source".  If I put "disable ntp" in the configuration file, does ntpd
actually continue to talk to the stratum 1 NTP servers I configured?

DLM> Run the machine for a day or so and use ntpq to see how the
DLM> offset diverges.

I looked for "offset" in the ntpq docs, and found it gets printed in
two places:  by running the internal "peers" command, which returns
"No association ID's returned", and by running "readvar", which always
prints an offset of 0.000.  Is "offset" the difference between my
system clock and some reference that ntpd knows about?

It seems to me that if "disable ntp" in the configuration file
actually disables contacting NTP servers, then there would be no way
to calculate an offset.

DLM> If the jitter.c program shows good and the offset looks clean and
DLM> straight, adjust the cycles_per_jiffy to correct the apparent
DLM> frequency offset.

Is there some documentation for the jitter.c program?  I'm not clear
on how to analyze the output I get:

  # ./jitter > gtod
  First rank
           0         0
           1         0
           2         0
           3         0
           4         0
           5         0
           6         0
           7         0
           8         0
           9         0
  Last rank
       19990         0
       19991         0
       19992         0
       19993         0
       19994         0
       19995         0
       19996         0
       19997         0
       19998     10000
       19999     10000
  # 

Dave



More information about the questions mailing list