[ntp:hackers] Re: magnitudes for ntpq variables

David L. Mills mills at udel.edu
Wed Mar 29 03:27:32 UTC 2006


gnu,

All time satistics produced by the mode-6 (ntpq) support are in 
milliseconds; all frequency statistics are in PPM. All poll intervals 
are in log-2 seconds. There is no ambiguity whatsoever about these 
things. This is an inherent characteristic of the mode-6 monitoring 
protocol; ntpq just displays what it gets.

Dave

gnu not unix wrote:

> Hi graphers-of-ntp-stats--
>
> I'm converting to ntpq to gather stats for rrd plotting.
>
> Compared to ntpdc, which uses nice scientific notation when
> dumping "-c kern" values, there is some ambiguity (in my mind
> anyways) about the stats out of ntpq, for these:
>
> offset, jitter, noise, stability
>
> The frequency out of ntpq and ntpdc is understood to be ppm,
> so no problem there.
>
> If the values are taken as-is, the graphs look rather coarse
> compared to using the ntpdc values (eg offset and jitter).
>
> Looking at the "ntpq -p" columns for offset and jitter tend
> to make me think that the similar "ntpq -c rv" variable names
> should be scaled by 1000--since the "-p" stats are "millisecond
> equals 1.0", not so?
>
> The web man pages did not clarify this issue, that I could find.
>
> Stability is mentioned on the web page for ntpq, but not noise.
> Am I sent to the corner with RFC-1305 in my hands?
>
> I'm wondering if I should also plot root delay and root dispersion.
>
> The ntpq -p delay column is interesting, as it seems to show some
> of the udp stack latency on my 100meg lan peers. I'm not sure what
> the equivalent is in the ntpq "-c rv" output, though.
>
> ../Steven




More information about the hackers mailing list