[LEAPSECS] LEAPSECS Digest, Vol 25, Issue 5

[David Malone]

> Does anybody have any thoughts on the effect(s) of the previous and/or

> upcoming leap seconds on NTP (v2, v3, and/or v4)?  For example, on

> protocols that implement sub-second timers (such as BFD)?  TCP?  IGMPv3?


I recorded the behaviour of a couple of different systems during
the last leap second:

	http://www.maths.tcd.ie/~dwmalone/time/leap2005.html

The bottom of the page has links to other people's observations.
I did hear a report of one protocol stalling because it saw an
out-of-order timestamp over the last leap second, but it was an
in-house protocol and I wasn't able to get details to actually
confirm the problem.

As Warner said, NTP should behave fine modulo bugs. Most low level
shouldn't need to convert from machine counters to real time, so
the leap seconds shouldn't have an impact. For example, TCP timestamps
in FreeBSD are implemented using the kernel "get*uptime" functions
and in other OSes they often seem to be implemented in terms of
interrupt ticks/jiffies/... Interestingly, this can actually offer
some chances for fingerprinting devices based on clock drift:

	http://www.caida.org/~yoshi/KoBrCl05PDF-hires.pdf

David.