[geeks] [JuliaSacks930 at hotmail.com: [janglo] INFO4U: My Ipod Broke My Computer]

[der Mouse]

>> Hard disks have an MTBF of about
> 300,000 hours

>> So if you go out and buy an iPod and connect it to your computer,
>> the odds of the hard disk dying within the first hour are about
> 1 in 300,000...

Not quite.

MTBF doesn't tell the whole story.  You're assuming that failures are
uniformly distributed, in a sense rather like radioactive decay: that
the chance of a drive dying during a particular time interval depends
on nothing but the size of that interval, whereas in reality it also
depends on how far into the drive's service life that interval falls.
If every drive failed at exactly 300,000 hours, MTBF would still be
300,000 hours, but the odds of a drive dying in its first hour of
service would be zero.  Alternatively, if half the drives die
immediately and half fail at 600,000, MTBF is still 300,000, but the
chance of an immediate death is 1/2.  (Of course, the reality is
neither of these; it nay not be uniform, but certainly is not as
sharply clustered as these examples.)

> But, for every drive that lasts 300,000 hours, doesn't that mean
> there is one that fails in the first hour, and another drive that
> lasts 600,000 hours?

Only under the assumption of uniform failure distribution, which does
not match reality.  Some of the failure modes - such as spindle
bearings wearing out - have a relatively well-defined service life,
with failures relatively clustered around that time.  Other failure
modes are more uniform, closer to the radioactive-decay model, such as
certain kinds of IC failure.  Still others exhibit a bathtub failure
curve (infant death plus old-age wear-out).  Mix all the various
failure modes and you get a failure time distribution which is
mathematically messy, but for which MTBF is still a useful "sum it all
up" indicator.

> Just an question, but can they really warrant the external contacts
> for that long?  (Said another way, do we know that corrosion won't
> erode the electrical properties of the contcts in less than 34
> years?)  The short answer I guess is no, that is why drives have 1-5
> year warranties ;^)

If the MTBF is 34 years, and they offer a 5-year warranty, then they
should expect, ummm... .5^(5/34) is about .903, so, about 10% of their
drives to fail in-warranty.  Under the assumption of uniform failures,
that is; if they do factory burn-in to eliminate the early deaths from
causes that exhibit bathtub failure curves, and the causes such as
mechanical wear-out fail well past the warranty period, warranty
failures will come predominantly from other causes, and will probably
be significantly lower than this calculation implies.

Anyone know of any public data on actual drive failure rates with age?
I imagine disk makers have such data, but probably don't publish it;
I'd be interested.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse at rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B