[geeks] Disks: recommendations?

[Nathan Raymond]

On Fri, Oct 30, 2020 at 11:34 AM Phil Stracchino <phils at caerllewys.net>
wrote:

> I think SSDs still tend to abruptly and completely fail.  Either they
> work or they don't, they don't really have a "degraded" mode.
>

They're actually wearing out constantly. Every write causes that part of
the SSD to wear out. Eventually an SSD will be unable to be written to (how
it behaves is somewhat unpredictable, and while many are supposed to go
into a read-only mode, when tested a lot of SSDs just seem to become
unusable when their lifespan is up). SSDs have a hidden reserved area and
they manage what is hidden and what is exposed internally and abstract it
from the host, and then use the hidden spare area to make sure that data is
written as evenly as possible across the SSD so that any part doesn't wear
out any faster than another (this is called wear levelling). Also writing
anything to an SSD involves erasing that block first. By having a hidden
area (as much as 30% of the total capacity in an enterprise SSD, less in a
consumer one) the drive can do background erasures of freed up blocks. The
TRIM command is what allows an OS to communicate to the drive that logical
blocks do not have filesystem data in them which allows the SSD to also
pre-erase those blocks outside of the hidden area which improves write
performance. If an SSD has to erase a block just ahead of a write operation
because the block wasn't already erased ahead of time then SSD write
performance will tank. And as you mention, there are cases where SSDs will
sometimes abruptly fail ahead of their expected failure point, and this is
usually due to firmware bugs because SSDs are rather complex from an
internal software perspective compared to hard drives.