Subject: | |
From: | |
Reply To: | |
Date: | Wed, 20 May 1998 09:46:59 -0700 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Wirt writes:
> ABC News is reporting another rollover date problem that's somewhat
> interesting. It occurs that the Global Positioning Satellite system encodes
> its week count as 10 bits (0 to 1023 being the possible numbers),
> beginning on the 5th of January, 1980.
I believe this is at least the second GPS field that's wrapped around,
so this kind of problem isn't entirely new to the GPS community.
It does underscore the worst-case nature of date sensitive bugs, as their
synchronous nature means that everyone hits the bug at the same time,
swamping whatever support services there are and preventing the customer
from simply switching to an identical backup system.
In the 3000 world, there have been several cases of this kind of problem
too. There was the recent COBOL runtime problem where it was multiplying
by part of the CALENDAR intrinsic result, overflowing a 16-bit field, and
I can remember a couple of January 1's where SLEEPER went into a hard loop,
starving every machine for CPU (not to mention failing to launch any
jobs).
G.
|
|
|