Hi jdow!

 On 2013.06.10 at 19:06:56 -0700, jdow wrote next:

> This is mentioned so infrequently in the literature that I am not sure it
> has been generally recognized or dealt with by the disk manufacturers. It
> surely astounded us when the reports started coming in.

This is pretty interesting, thanks. I heard mention of it "in theory",
but it's useful to know that it can happen in practice.

This is really unobvious because generally manufacturers warn only about
amount of erase cycles as limiting factor. But I've read some
calculations which have shown that theoretical "pure erase" cycles count
is much higher than what SSDs are rated for - but in real life, when you
do lots of reads too, you can think of rated "erase cycles" count as of
actual value.

People do crazy amount of writes (i.e. way more erase cycles than drive
was rated for) when they do synthetic experiments like
http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm&p=5132834&viewfull=1#post5132834
(4 PB written on consumer MLC SSD; even Intel X25-E isn't rated for that
much!)
But some people explain in thread there that from NAND perspective, such
high amount of erase cycles doesn't mean much when data is just written,
never read, and if it was read, that SSD would've been already long
dead.

I suppose that by that elaborating that logic, one can assume that if
you do very high amount of reads from certain NAND, it's supposed to
fail before reaching rated amount of erase cycles.. Well, you confirmed
that it can be the case.

Proves that rated "erase cycles" count is pretty abstract and has
meaning only for some "typical usage pattern" - cases where you don't
have to worry about their amount anyway. And for specific use cases one
really has to look at expensive SLC or high-endurance MLC models, or at
least carefully evaluate various factors..

-- 

Vladimir