In message <4302512F.9040500@ntp.org>, Frank Kardel writes:
>Hi *,
>
>I am running 3.99.7. During the tests for the next ntpd release I noticed
>following behaviour on the NetBSD ntpd and the ntp-dev version:
>
>1) Configure ntpd to have peers in the internet reachable via pppoe0.
>2) Start ntpd.
>3) All is well.
>4) Do ifconfig pppoe0 down; ifconfig pppoe0 up.
>5) See ntpd still sends from the old address (tcpdump -i pppoe0 port 123).
>Unsurprisingly replies are not received.
>
>When you re-start ntpd all is well again.
>
>Am I the only one seeing this ? This does not seem right as the
>previous address is gone it doesn't seem to make sense that udp
>endpoints remain stuck on invalid addresses. Not even an error is
>reported in ntpd. Applications seem to be kept in the dark about
>being disconnected from the sane world.
>

It's not pppoe; it's any interface.  If you run netstat, you'll see 
that ntp explicitly binds to every IP address it sees.  As a result, my 
dhclient-exit-hooks restarts ntpd if addresses have changed.

		--Steven M. Bellovin, http://www.cs.columbia.edu/~smb