[nylug-talk] Paper IT certs and disk drive fabrication differences -- WAS: Slim home server

[Eric Moore]

"Bryan J. Smith" <b.j.smith at ieee.org> writes:

> But MTBF is not considered an usable number though, and hasn't been
> for a long time.  But it is still quoted.  A better practice is the
> change of failure, assuming standard operating conditions, per 6 or
> 12 months.  It currently stands at just over 2%.  Different lots
> often vary by +/-0.6% or greater last time I checked, but I'm
> probably outta date.

If you're modelling it as a constant chance of failure for some block
of time, you should be able to calculate a MTBF.  The exponential
probability distribution has a mean, and if the failure rate is
constant, the time before failure will be exponentially distributed.
2% per 12 months would give a MTBF of 49.5 years, or 433900 hours.

As long as you're assuming the failure rate is constant, the MTBF and
the chance of failure are equivalent numbers.  If they're quoted in
the same units, MTBF = 1/ln(1-rate).  How can one be more usable than
the other?

-- 
Eric