Something is buggy with iburst...

[Achim Gratz]

Hal Murray via devel writes:
> Do you have rawstats enabled on the client?
> Is there more then one client?

I have four NTP servers on the same switch, two have GPS/PPS and are
used as primary servers.  The other two have a serial or USB GPS that is
currently configured "noselect" and are locking to the primaries.  They
look at each other and the same set of external servers for monitoring
purposes.

> Are the servers yours too, or something out on the net?

See above.

> The default rate-limit is supposed to work with an iburst.  It might complain 
> if you restart a server too rapidly - the server will still remember the 
> previous iburst when the new one comes along.

I have iburst switched off at the moment, which has helped the startup
problems, but not the occasional glitch.  All four servers are within
200…600µs of network delay, which may be a factor.  Before I started
noticing these problems I've had the (at the time single) primary server
on another port that would have the packets go through another switch
which probably added another couple dozen µs delay.

> You can also get in troubles if you are going through a NAT box and have 
> several servers using the pool that get restarted at the same time.  They get 
> the same answers from the DNS and the server sees several ibursts at the same 
> time when NAT maps them all to the same IP Address.

I've only had trouble with the external servers one single time and I
believe that was caused by me restarting all four boxes several times
in short succession (I've had iburst switched on then).


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Wavetables for the Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#BlofeldUserWavetables