[ntp:questions] Re: NTP Synch >125msec Network Delays?

Brian Utterback Brian.Utterback at Sun.removeme.COM
Wed Nov 12 14:39:00 UTC 2003

B. Dairiki wrote:
> Clarification on earlier post.  Thanks for response
> Brian.
> Just to clarify.  I'm interested in network delays of
>>128 ms and its impact on NTP synch.
> Suppose I have a NTP time distribution local
> configuration (IRIG-B reference - stratum 0, several
> NTP stratum 1 servers, and multiple NTP stratum 2
> servers) and I'm interested in linking up a few
> distributed (very far away) workstations (NTP
> stratum 2) over some long haul network that may have
> network delays of say 300-500 ms (one-way) and assume
> the round-trip is symmetric.  For these distributed
> workstations, there is no NTP stratum 0/1 source other
> than across the long haul network.
> Can I expect NTP to maintain time synch on these
> globally distributed workstations with high network
> delays?

Didn't I just say that? The length of the delay doesn't matter
at all, as long as it isn't so long that NTP has forgotten that
it asked the question before the answer came back. The symmetry
of the connection is the key here.

That having been said, the delay does have an effect in the
overall accuracy. You cannot get perfect symmetry, there is
always some error. The amount of the error is bounded by the
total delay. We know what the offset between the server and
the client is by simply subtracting the timestamps. The client
knows that the interval defined by the timestamps provided by
the server is completely contained in the interval defined
by the two timestamps that the client itself generated (in
other words, the server had to receive the request after the
client sent it, and the server had to send the reply before
the client received it). However, the client cannot know
where in the larger interval the smaller actually lies. It
assumes that the interval is exactly in the middle. How far
off this assumption is defines the error.

So, the error is cannot possibly be larger than the larger
interval, so the smaller that interval is (tiei.e.. the smaller
the delay) the smaller the largest possible error is. But
since the error is really introduced by the actual asymmetry,
in practice it is much smaller. The real error is more
often caused by the jitter introduced at each hop. This jitter
contributes to the asymmetry and thus to the actual error.


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