[LEAPSECS] Focus in the debate, alternative proposal

[Warner Losh]

On 01/07/2011 09:16, Paul Sheer wrote:

> On Fri, 2011-01-07 at 13:08 +0000, Zefram wrote:

>> being a second out of synch would cause reconciliation problems.

>

> can you explain this closer to the tech please?

>

>

> i know i got the same explanation from one of our support

> people until I challenged him...


Both sides of the transaction stamp the transaction with timestamps with 
microsecond resolution.  When the trades are reconciled at the end of 
the day, these timestamps need to match with a fuzz factor that's in the 
sub-millisecond range and getting smaller as the years go by.  Otherwise 
the trades need to be reconciled by hand as being anomalous.  The 
trading operations have some fairly strict timing requirements to ensure 
that everybody is operating legitimately and there's no fraud.  If there 
were a significant portion of the day where the time stamps were off by 
a whole second, then reconciling thousands of trades by hand would be 
burdensome and cost prohibitive and would interfere with the free flow 
of the markets.


> we then went into the code and analysed what would happen

> on one of our live systems and discovered that the only

> two problems we would ever have was the *appearance* of

> wrong timestamps to someone eye-balling the logs, and

> a window of bad performance stats in cases where people

> were doing cross-system performance profiling.

>

> But that otherwise, no-one would notice.

>

> For this reason, I am hesitant to believe the argument:

> "millisecond time stamps are used and *therefore* leap

> seconds cause havoc."


While your caution here is admirable, time skews do cause actual 
problems in other systems.  But be careful here: just because some 
systems it doesn't matter doesn't translate to all systems not mattering.

Warner