A leap second was inserted into the last minute of 2005. This page contains some graphs showing how various NTP peers of ours behaved during that leap second. I have also collected some other leap second observations.
The graph below shows a well behaved peer during the leap second.
The X-axes shows the number of seconds to the leap second, so the
leap second is where X=0. The Y-axes has units of milliseconds and
shows the offset, delay and jitter as reported by
-p. All values are small, and the offset is quite smooth
apart from occasional jumps that seem to correspond to jumps in the
delay (round trip time to the remote machine).
We can also zoom in around the leap second to see the detail. The small gap visible just before the leap second is a missing response from that server.
There are some peers who behave well, but you can see events that are not leap second related. For example in the following graph we see a shift in round trip time (probably due to a routing change) which results in an adjustment in the offset.
Around the leap second we see the peer is well behaved.
This machine seems to have mishandeled the leap second resulting in a big spike in the offset just after the leap second. This is followed by a step, to correct the clock and then the offset decays to 0 as over time.
Zooming in, we see this more clearly. I'm not sure why the offset is 0.5s just after the leap second - I would have expected it to be 1.