On 2012-06-26, Howard Brazee <
how...@brazee.net> wrote:
> On Tue, 26 Jun 2012 06:01:25 -0400,
jack...@bright.net wrote:
>
>>I've been lucky enough not to be on duty during any daylight savings
>>time shift, because I don't know how the bosses would react to the log
>>entries I would be compelled to make:
>>
>>01:59 Entering time warp
>>01:01 Exiting time warp
>
> That can be expensive in data processing. I've worked with time
> sensitive databases that took a couple of hours to adjust 1 hour, so
> that sequencing did not change.
>
> They never should have been stored with local time stamps anyway. Zulu
> works better.
I've been bitten by this going both ways.
I was an operator at a small stock/securities exchange for a while.
Expected to follow local hours, open at 9 am and close at 5:30 pm. But
the hours for the various states the markets could be in were stored in
Zulu, so we had to go in and tweak the state transition timers by two
hours in the database twice a year. (Needless to say this could've been
handled automatically, but that was never a development priority during
my time there...)
Then, another transaction-intensive system, but open 24/7 to people in
various time zones. This one stored events in local time, so we had
planned downtime around the DST changes. The stated scenario they were
afraid of was two events ending up with identical timestamps, which
would have unpredictable effects. Of course, that doesn't explain why the
system also had to be down when the clock moved forward, but the orders
were clear, even after we tried to reason with their originators...
So in short, I guess this is a design decision that requires careful
thought. Much like many others.
Niklas
--
if (!this->with_sin())
(int) stones[0];
-- Joe Pfeiffer's friend