[LEAPSECS] one second tolerance

Warner Losh imp at bsdimp.com
Wed Feb 9 11:15:05 EST 2011

On 02/08/2011 23:54, Mark Calabretta wrote:

> On Wed 2011/02/09 06:25:25 -0000, Ian Batten wrote

> in a message to: Tom Van Baak<tvb at leapsecond.com>,

> Leap Second Discussion List<leapsecs at leapsecond.com>


>> Although that mandates access to a continuously reliable source of DST

>> changeover dates and offsets. It also opens the interesting question

>> of what timestamps mean in a non-monotonic timescale: during the autumn

>> change, a hour is repeated. So as far as a clock ticking civil time is

>> concerned, you have 0100 through 0200 twice, once a year. Events

>> stamped 0130.01 and 0130.02 could be an hour and a second apart.

> Not if the timestamps are properly labelled with the timezone,

> preferably specified as an offset, which distinguishes between

> DST and non-DST.

I think the view is that there's no trading going on in the missing hour
as the exchanges are closed then. Whether this is, in fact, true in the
future remains to be seen.

After seeing this last night, I also contacted some friends that I have
where I helped setup a timing system that I talked here about here a
while ago. I asked them about this and their insistence to me about the
millisecond level timestamps. There were two reasons for them. (1)
they use them as internal controls for auditing (which is what I'd
confused for external controls: they match them against their own
systems which are sync to the submillisecond level so large deltas
between the indicate problems) and (2) they are used by their algorithms
for future trading purposes. My friend was surprised that they didn't
mandate tighter requirements, since most traders would easily be able to
comply (since most trading firms already have highly synchronized clocks
which helps to drive some of their trading algorithms). The trading
aspect of time drove the requirements, since from it they can make more
money than they can without synchronized clocks.


More information about the LEAPSECS mailing list