Re: MTBF interpretations (Re: ZFS performance)
hede wrote:
> On Fri, 11 Nov 2022 14:05:33 -0500 Dan Ritter <dsr@randomstring.org> wrote:
>
> > Claimed MTBF: 1 million hours. Believe it or not, this is par
> > for the course for high-end disks.
> >
> > 24 hours a day, 365 days a year: 8760 hours per year.
> > 1000000/8760 = 114 years.
> >
> > So, no: MTBF numbers must be presumed to be malicious lies.
>
> With your interpretation every single drive would not be allowed to fail before its MTBF value. That's wrong. MTBF is a mean value for all drives of this type, not a guaranteed minimum value for a single drive.
No, my interpretation is that the average (mean) lifetime
between failures should be the listed value. At 114 years, half
of the population of drives should still be working.
This is obviously not congruent with reality.
> If you have one thousand drives then you have to expect one drive failure for every one thousand hours (MTBF=1,000,000). Statistically you have to expect nearly nine of those drives to fail per year (if I did my math correctly). With higher MTBF values this value will be smaller, with lower MTBF values you'd have to expect more drives to fail.
You are assuming a linear distribution. A normal distribution
would be more likely. A bathtub curve is often hypothesized.
The largest public source of data on disk mortality is from
BackBlaze:
https://www.backblaze.com/blog/backblaze-drive-stats-for-q3-2022/
and what you can work out from their data set is:
* Brands are not meaningful as predictors of longevity
* It is not linearly distributed.
* It is not normally distributed.
* It is not a bathtub distribution.
Some drive models are lucky. Some are unlucky. Overall you
should expect about 1.54% of disks to fail each year in a large
mixed-age population, not quite double what you estimated. But
the range of annualized failure rates is from 0 to 9% -- you
could be lucky, or very unlucky.
-dsr-
Reply to: