[svlug] More links about ssds

Rick Moen rick at linuxmafia.com
Thu Jun 2 09:48:27 PDT 2016

Quoting Marc MERLIN (marc_news at merlins.org):

> Ok, so at least we have this guy who tested the old generation:
> http://techreport.com/review/27062/the-ssd-endurance-experiment-only-two-remain-after-1-5pb
> http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

At a quick look, that's really good, thanks!

> If we say about 1PB for the old 840 246GB, that's at least 8PB for the 2TB
> and assuming the 850 is better than the 840, you ought to be able to count
> on at least 10PB

Well, we can _hope_ the 850 Pro is better than the 840 Pro, and we might
even hope that the 850 _EVO_ is better than the 840 Pro -- but the most
obvious difference between the 850 EVO (which you were discussing
upthread) and the other two Samsung models is that it relies primarily
on triple-level cell (TLC) NAND flash.  The other two use longer-lived,
more expensive multi-level cell (MLC) NAND flash.

Which is probably also why the 850 Evo is $90 versus the 850 Pro being
$127:  same capacity, same form factor, but the Evo is cheaper because
on average it'll fail sooner than the Pro, because of its TLC reliance.

I'm not arguing for any one of these units over the others -- and, when
I _did_ buy a pair of SSDs for my new server a couple of years ago, I
bought the then-sweet-spot Samsung model.  However, personally, if
choosing today among the two you discuss, I'd pay the extra $37 --
because, much as I like the bucket of steamed clams in herb de provence,
Chablis, onions, and butter, with garlic bread, and half-pitcher of
beer at The Dutch Goose that the $37 buys me, I value the time not spent
RMAing failed mass storage just a bit more.

Just to circle back:  My upthread simple disambiguation of points of
comparison between SLC, MLC, eMLC, and TLC was _not_ an assertion that
it's 'that simple' -- because few things are ever 'that simple'.  It was
a 'all other things being equal' point of comparison among those four 
NAND classes _generically_.

The average endurance of a particular model Samsung (for example)
manufactures will also depend on all the _other_ engineering the firm
sinks into it.  In which case, all other things are likely to not be

Notwithstanding which, the savings equal to the cost of a lovely dinner
at The Dutch Goose suggests to me probable shorter life, on average.

> What's the write endurance of a recent 6TB spinning rust drive?

One of the _other_ many points I made during my lecture is that the
biggest single reason SSD wear is a red herring is that your SSD is
already very highly likely to outlast the HD you buy at the same time.

Sarah made the interesting point that she's seen SSDs fail
catastrophically with very little advance indication of impending doom,
which might take one by surprise if one is used to HDs that _most_ often
signaling frantically that they're on their last legs and you should be
doing that final backup instanter.

[SMART data:]

> Haha :)
> Some do, some don't. Some data on some drives is actually pretty accurate
> and useful.

But manufacturers are notorious for making the SMART data say what they
want it to, with the consequence that it's the best indication of drive
health and remaining life only because it's the only indicator we have.


More information about the svlug mailing list