[ntp:questions] Network problems affecting NTP

David Woolley david at ex.djwhome.demon.invalid
Fri Jul 1 07:03:51 UTC 2011


Eugen COCA wrote:

> One may observe the delays were also affected during this interval.
> Also, the time information served to clients was affected, the
> pool.ntp.org monitoring system confirmed the delay -
> http://www.pool.ntp.org/scores/80.96.120.252. It is obvious that a
> server synchronizing only with Internet sources will be in error as
> there is has no way to determine the problem. Can someone explain this
> behavior ?

What do you mean by wrong offsets?  How do you know that the offsets are 
not more correct in the fault configuration?  OK you are assuming that 
the PPS is valid, although even that might have been calibrated on wrong 
assumptions about the time at the point of calibration.

You have some very bad choices of server (140ms delay).  Your external 
network connection has very high latency, by modern standards, even in 
the non-fault situation.

If you have a round trip delay of n ms, NTP assumes that the upstream 
clock was sampled n/2 ms before its response arrives, however, subject 
to any minimum delays, it could be anywhere between +/-n/2 mas relative 
to that.  A longer network route increases the chances of this sort of 
asymmetry.

tinker huff and puff provides some protection against more severe cases, 
but cannot deal with systematic asymmetries.  You don't seem to have a 
particularly severe problem - your offsets are quite good given that 
most of your servers are too far away in network terms.




More information about the questions mailing list