[ntp:questions] Re: Behavior when one server is erroneous

David L. Mills mills at udel.edu
Sat Jan 17 02:37:42 UTC 2004


Christopher,

There is a very sophisticated set of algorithms to wrangle a possibly
unruly server herd. See the briefings at the NTP project page linked
from www.ntp.org and, in particular, the architecture briefing, which
describes how these algorithms work. Sometimes it may seem
counterintuitive that these algorithms behave as they do, but understand
how they work before concluding they are doing the wrong things.

Dave

"Christopher J. Burian" wrote:
> 
> I'm testing a CDMA-based, Palisade-compatible time source.  Most
> of the time it works fine.  But in marginal reception conditions,
> the device thinks it is synchronized but is not, and appears to
> be off by up to a second.  I'm working with the manufacturer on
> that problem.
> 
> What I wanted to bring to the attention of ntpd developers is
> that the daemon appears to do some sort of weighted averaging
> of the three mostly correct servers and the one bad server.
> Shouldn't the algorithm perhaps vote to reject servers notably
> different from consensus?  And when jitter is bad, shouldn't
> the polling time be more frequent?
> 
> Regards,
> Chris
> 
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset  jitter
> ==============================================================================
> *127.127.29.1    .CDMA.           0 l    7   16  377    0.000  -26.916 116.099
> +155.34.100.67   .GPS.            1 u  184  512   37    3.277  450.174  80.107
> +155.34.3.123    .GPS.            1 u  185  512   37    5.223  450.440  79.103
> +155.34.3.8      155.34.3.123     2 u   23   64  377    0.002  574.188  25.114
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset  jitter
> ==============================================================================
> x127.127.29.1    .CDMA.           0 l   10   16  377    0.000  -226.84  99.691
> x155.34.100.67   .GPS.            1 u  331  512   37    3.277  450.174  80.107
> x155.34.3.123    .GPS.            1 u  332  512   37    5.223  450.440  79.103
> *155.34.3.8      155.34.3.123     2 u    5   64  377    0.002  583.724  14.496
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset  jitter
> ==============================================================================
> *127.127.29.1    .CDMA.           0 l   16   16  377    0.000  -671.47 141.500
> x155.34.100.67   .GPS.            1 u   22  256   17    3.357  -74.293  46.467
>  155.34.3.123    .GPS.            1 u   21  256   17    5.096  -19.350  72.238
>  155.34.3.8      155.34.3.123     2 u   56   64   76    0.919  -21.359 177.973
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset  jitter
> ==============================================================================
> *127.127.29.1    .CDMA.           0 l    5   16   37    0.000  -75.688  41.494
>  155.34.100.67   .GPS.            1 u   20   64    1    9.419  750.087   0.002
>  155.34.3.123    .GPS.            1 u   21   64    1    5.276  746.882   0.002
>  155.34.3.8      155.34.3.123     2 u   20   64    3    0.002  288.527 457.307
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset  jitter
> ==============================================================================
> *127.127.29.1    .CDMA.           0 l   11   16  377    0.000  295.474  70.632
> +155.34.100.67   .GPS.            1 u  183  512   37    3.707  784.915 144.207
> +155.34.3.123    .GPS.            1 u  183  512   37    5.202  897.152 125.242
> +155.34.3.8      155.34.3.123     2 u   25   64  377    0.002  1129.50 168.278



More information about the questions mailing list