monitoring, was Re: [ntp:hackers] Re: Testing a CDMA based time server

David L. Mills mills at udel.edu
Thu Jan 15 10:09:00 PST 2004


gnu,

Why use ntpdc? If all you are doing is watching offset, etc., ntpq is
much better and more reliable than ntpdc. The ntpdc has not been
faithfully maintained to match NTPv4 statistics, so things like jitter,
dispersion, stability and related system variables may not be correctly
reported.

Dave

gnu not unix wrote:
> 
> In message <4006C9DB.51862C62 at udel.edu> you write:
> 
> >I can't say much about accuracy other than noting the residual offset
> >and jitter displayed by ntpq, one microsecond offset, 15 microseconds
> >jitter. The claimed precision (-16) is only to 15 microseconds; I would
> >think the Linux kernel timekeeping is better than that. I see nominal
> 
> The linux kernel is being reported by ntpq -rv as:
> 
> system="Linux2.2.13-DOC"
> 
> This is rather old at this point.
> 
> By the way, I've set up a web page that plots error, offset and
> frequency as reported by ntpdc -c kern for some of the pool.ntp.org
> hosts. Its been quite interesting to watch the various bumps and
> crashes over the day or so that the plotting covers.
> 
> Most of the error plots seem to follow the spike/tail one would expect
> of the ntpd over a bumpy net. At least one unit shows an odd staircase
> pattern. And one of the pool units seems to be one of those
> CDMA servers running 2.2.13.
> 
> http://www.wraith.sf.ca.us/ntp/pool/index.html
> 
> I also plot my experimental units at:
> 
> http://www.wraith.sf.ca.us/ntp/rrd/index.html
> 
> If anyone would like to be added to the plot list, send
> an email with the IP address you'd like monitored.
> Since the script uses ntpdc mode seven, its likely that only
> true-blue ntpd will be able to be plotted.
> 
> ../Steven



More information about the hackers mailing list