On Tue, 2014-01-21 at 08:31 -0500, The Wanderer wrote:
> On 01/20/2014 09:34 PM, Paul Wise wrote:
>
> > On Tue, Jan 21, 2014 at 1:22 AM, Holger Levsen wrote:
> >
> >> wait, what? Do you have any vendor statements to support this 20%
> >> extra space?
> >
> > Flash is basically probabilistic storage and you need extra space to
> > ensure that the probability your data is stored remains high,
>
> That doesn't match my understanding, and I don't see anything to support
> it in the reference links you gave.
>
> My understanding is that flash is unreliable in that individual flash
> cells will die (cease to retain what is written) after "too many"
> write/erase cycles, where the value of "too many" goes down as the
> fabrication process gets smaller. To mitigate the real-world effect of
> this and extend drive life, manufacturers build in more flash than the
> reported drive capacity, a practice known as "overprovisioning".
> Modifying the drive firmware can make it possible to access this flash
> as normal storage, at the cost of drive lifetime.
>
> The reference links you provided do state that flash memory is
> "inherently unreliable and unpredictably fragmented", and one of them
> also states that a large fraction of the memory on any given flash die
> is likely to be bad sectors, but nothing I've noticed in them supports
> the claim that there's anything probabilistic about it.
[...]
Even 'good' sectors of NAND flash can have many errors, which is why
complex ECCs are used. Probably you will get the right bits back - if
you haven't rewritten it too many times.
But really, every storage and communication medium is subject to errors
and manufacturers must do some statistical analysis to validate that the
uncorrectable error rate is not too high. NAND flash just has an
unusually high 'raw' error rate.
Ben.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Attachment:
signature.asc
Description: This is a digitally signed message part