On Mar 19, 2013, at 10:07 PM, Fred Cisin <cisin at xenosoft.com> wrote:
On Tue, 19 Mar 2013, David Riley wrote:
They can demonstrate given empirical evidence.
we determine the MTBF of electronic devices, it all comes down to
statistics. If you test a million capacitors over a period of a
year and five of them fail, statistically a given capacitor will
fail every 200,000 years.
IFF you postulate and accept that the failure rate is linear.
(I have no idea whether it IS)
That's why you have to make sure your metrics make sense. In
MTBF analysis, there's a "bathtub" curve with the "infant
mortality" at the beginning and the eventual inevitable demise
of all devices at the end. Burn-in testing of products is
supposed to eliminate the former from the reliability equation,
so in theory you're left with mostly the flat part of the curve
as well as the tail end.
Consider: of a million cartons of milk, 5 go bad in a
those cartons of milk are good for eight hundred years
Cartons of milk have a relatively short time before you hit the
tail end. 5 go bad in a week, maybe, but that's not doing
nearly enough testing to discover the point of 100% failure.
If you go another week out, you'd get better data; a month out,
even better. If you accelerate it with heat (up to a certain
point, obviously), you can do more testing in less time if you
have studied the mechanics of how the heat changes the failure
rate (it's obviously not necessarily linear, but within a
certain range it follows a known correlation).
If 4 out of a million of these discs fail in a week,
then they are good
for 1000 years?
If you've established that a week is long enough to test it
under appropriate environmental conditions, yes. In most
cases, it's not, and there's a LOT of research that tells you
what conditions you need to test under and for how long.
And again, you can blithely ignore that research, say you've
"tested it", and people will buy it on that. I haven't the
slightest clue what the case is here, but if I were thinking
about using these things for archival media, I'd certainly
be trying to establish their bona fides.