[ntp:questions] Re: Dopey selection of server

Brian Utterback Brian.Utterback at Sun.removeme.COM
Fri Jan 30 18:53:11 UTC 2004


Christopher Burian wrote:
> Can anyone explain this:
> 
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset 
> jitter
> ==============================================================================
> +cs.columbia.edu 204.123.2.5      2 u  725 1024  377   16.819   -7.688  
> 2.119
> +filbert.cc.colu 128.59.39.48     2 u  704 1024  377   22.523   -4.356  
> 5.053
> *cudns.cit.corne 192.5.41.40      2 u  656 1024  377   60.335   13.325 
> 23.390
> ntpq>
> 
> Instead of rejecting the one that is off the most from the others, it is 
> selecting that wayward server, despite it having the worst jitter as well.

In addition to the RV billboard data, another thing that you missed is that
NTP works with error intervals, so a particular candidate is examined in
the interval offset+/-jitter. A consequence of this is that it takes two
servers to define a candidate interval. This means that for NTP to decide
that any server is not as good as a bunch of others, there must be at least
*three* that agree. In this case, you basically use the two that are in
agreement to define the interval, with nobody left over to vote for it.

It is therefore the case that you should have at least 4 servers, if you
want to insure that a bad server gets voted off the island.

-- 
blu

Lesson from the blackout of 2003:
The power grid is THE most critical infrastructure, upon which all
others depend, and nobody really knows how it works.
--------------------------------------------------------------------------------
Brian Utterback - Solaris Sustaining (NFS/Naming) - Sun Microsystems Inc.,
Ph/VM: 781-442-1343, Em:brian.utterback-at-ess-you-enn-dot-kom




More information about the questions mailing list