On 4 Sep 2010, at 19:17, Poul-Henning Kamp wrote:
In message <1283623216.9536.35.ca...@localhost>, Paul Sheer writes:
the software side I don't need to ask about - I've
been working on their core C code for two years now.
And...
it works perfectly even if the time is completely wrong.
Why? because none of the code depends, to-the-second, on
the absolute time of day.
And because your code is time-agnostic, you are 100% sure that no
telco anywhere has any leap-second issues with any code they run
in their system ?
Well, I have spent some years working on software for telcos,
including billing, management and call processing, with roles covering
coding, specifying and negotiating in amongst which I became the
"time, NTP and MSF" guy, and I never heard leap seconds mentioned by
the telcos (it never appeared in specifications or contracts, for
example).
I did raise the issue in one meeting and the view was that if it
happens on 31 Dec, the calling patterns around midnight (in the UK,
obviously, 23:59:60 UTC is the actual bells and streamers midnight)
are so atypical no-one cares about statistical consistency, and June
30, well the odd oddity, who cares?
I'm struggling to think of cases where it would matter to any extent.
Call processing relies on time stamps and interval times, but the
relationship with reality is irrelevant (the protocols are elderly,
and the idea that adjacent operators have sub-second agreement on time
over the life of the protocols is a stretch) and so long as the timing
is monotonic, I doubt it matters much.
One that might is billing If 0000UTC is a tariff increase point in
your timezone, say 8am in parts of Australia and China, then customers
may be overcharged for the last second of the interval. Someone's
bill might be inflated by at least one US cent several times a decade.
One particular paranoid installation shuts production, takes UPS
off-grid and runs on diesel during leap-seconds.
Is there the slightest evidence that is rational?
ian
_______________________________________________
LEAPSECS mailing list
LEAPSECS@leapsecond.com
http://six.pairlist.net/mailman/listinfo/leapsecs
