[ntp:questions] A Suggestion For Abolishing the Leap Second

David Woolley david at djwhome.demon.co.uk
Tue Jun 5 07:01:16 UTC 2007


In article <1181017458.302921.71210 at n15g2000prd.googlegroups.com>,
Quadibloc <jsavard at ecn.ab.ca> wrote:

> Managing the Internet:
> - A leap second means that millions of network router operators, in
> addition to periodically referencing their clocks to time signals,
> must also explicitly make an adjustment for the leap second at the
> proper time to forestall problems.

I don't understand this one.  Routers don't need accurate time in order
to route, and NTP is quite capable of signalling the leap seconds with no
operator intervention.

My impression is that it is the financial services industry, and associated
legislators, that have the problem, in that they want levels of simultaneity
that approach the limits of current technology, whilst being run by people
who are not scientists and therefore don't understand the reasons for 
civil time/TAI differences and expect to have their times in civil time.

Given that 1ms ~= 200km cable delay, it wouldn't surprise me if they
eventually reach a point where 1ms steps are unacceptable for trading
within a single city, simply because it is possible.

Databases that rely on monotonic time are a problem, but mainly becuase
coders don't know about leap seconds and operating systems don't provide 
access to TAI.  Historically discrepancies would probably have been allowed
to accumulate until the next reboot.

These issues would arise for SNA networks, not just for TCP/IP ones.  They
would even apply for a fully connected network, which wouldn't need an
internet in the more general, and original, sense.




More information about the questions mailing list