On 8/8/2018 10:54 PM, VanguardLH wrote:
mike wrote:

How's the reliability [of SSDs]? I'm still reading that they fail
catastrophically without warning.

Most have an estimated lifespan of 10 years, maybe more. For some
users, that's a lot longer then they plan on ever owning a particular
computer. When I build mine, I plan on a 6-year lifespan. My last
build was from a salvaged PC built in 2009 but I replaced a lot of
components (to repair the PC and to upgrade it) back in 2013. So it's
getting close to my 6-year expected lifespan; however, I'm disappointed
with the lack of progress in CPU, the Spectre problems, and will wait
another couple years before planning a new build. I added the SSD in
2016, so I should have several more years left before it gots kaboom.

Electronics seem to fail quick (in a month) or at about their pregnancy
period (9 months). Then they typically last until one month after the
warranty expires or, if you're lucky, until the device's MTBF. That
applies to HDDs or SSDs.

Yes, they do fail catastrophically. They suffer oxide stress at their
junctions during writes. All SSDs have a rated maximum number of writes
(but getting that info from the manufacturer is very difficult). I can
buy HDDs and SSDs with 5-year warranties. Don't plan on either of those
lasting much longer than the warranty. Anything longer is gravy.

They use wear leveling to increase their lifespan. They have to.

https://en.wikipedia.org/wiki/Wear_leveling

To compensate for when memory blocks go bad (and they will), there are
reserve blocks. Iffy blocks get their data moved (mapped) to a reserve
block and the iffy block gets marked as bad (so it never gets reused)
and any access to it get redirected to the reserve block. As mapping
increases, the drive slows down due to all the redirections. The
reserve space is limited. Once it is all consumed, the SSD drive
catastrophically fails.

That's the part I have issue with.
Just thinking out loud here...I'll not pretend to understand.

Sounds like some arbitrary architectural decision.

What if, when the spare pool ran out of spares, some additional spare blocks
got reassigned from one or more of the partitions.
The master file table got fixed up and the OS was informed
of the change? Maybe it's a drive firmware operation...maybe
it's an OS application/utility.

What is the difference between a reserve block and a regular
unused block? Can't a redirect point anywhere on the drive?

Why should a drive slow down because of redirections?
A pointer is a pointer. Shouldn't matter what the number is.

As you state below, the drive slows down because a densely
packed drive has to do read/modify/write on a whole block just to
save a small amount of data. Writing to a free block is faster.

On the surface, it seems crazy to have the drive just fail catastrophically
when some arbitrary number of reserved sectors gets used up, even
though the drive may have a huge amount of free space available.

When the free drive space on the partition gets low enough, writes
would fail for insufficient space and the OS already knows how to
handle that. The whole system shouldn't crash and
render the drive unreadable for recovery operation.

SSDs will get increasingly slower if the same memory block gets reused.
It must be erased (written) before it can be written with data. To
overcome this speed crippling, TRIM is used to perform optimization in
the background (when the PC is idle). Although the OS usually issues
the TRIM command, most SSDs nowadays have their own firmware do the
garbage collection during idle periods.

An HDD could die tomorrow and the SSD lasts for 10 years. An SSD could
die in the new few minutes while an HDD runs for 20 years. Depends on
how you use it, how it gets abused (by you, by surges, outages, temp
changes, etc), and the quality of the build. Faster with non-moving
parts does not guarantee greater longevity. Having non-moving parts
only imparts some probability of greater longevity; however, SSDs are
self-destructive, so they WILL DIE. Do *not* use SSDs for backups or
long-term archival storage unless the device is only used when saving
the data and not used to access it (except for restoration).

HDDs will fail, too. However, often they indicate pending problems via
S.M.A.R.T. data although that is not a guarantee. A drive whose SMART
says it is healthy could suddenly die, or get so flaky that they become
unreliable. One of the SMART attributes is a pending reallocation
count. HDDs also have reserve space to which bad sectors get remapped.
When a sector is flagged, the pending count goes up. When it eventually
gets remapped, the count goes down. The count should be zero. If the
count is not zero and doesn't eventually decrement to zero (i.e., the
flagged sector is not getting remapped) then the reserve space has been
consumed. The drive does not stopped functioning. There is no
catastrophic failure due to lack of reserve space. The pending count
keeps going up and the sectors become unreliable that cannot get
remapped. There are tools to watch the pending count (SMART monitors)
to warn you when this is starting, so you get time to recover or backup
as much as possible. When the SSD catastrophically fails, you'll have
to rely on your last backup whenever that was (and why backups should be
scheduled because humans make unreliable schedulers).

A lot of folks think that no moving parts mandates the SSDs will last
longer than HDDs. Wrong. An SSD that experiences lots of writes will
die sooner which could be much shorter than for an HDD with its moving
parts. HDDs will experience gyroscopic effects, have a maximum G force
they can withstand, and have a smaller operating temperature range.
HDDs are also not sealed, so they are affected by high humidity (water
molecules are smaller than the sinter filter in the body of the HDD).
SSDs can operate in harsher environments. SSDs do wear out.

SSD reliability in the real world: Google's experience
https://www.zdnet.com/article/ssd-re...es-experience/

I picked up a NEW Samsung SSD850 EVO 500GB drive last month at an estate
sale for $10. Not had the motivation to put it into anything.
Already forgot that I have it.
I'd still have to run the spinner for bulk storage.

I rarely reboot my system and don't play games.

I leave mine on 24x7 because I use my home PC at varying times: day,
night, early morning, any time. Plus I schedule jobs to run during the
early morn, so the computer must be ready (and putting it into Sleep or
Hibernate will only last about an hour before a scheduled task wakes up
the PC). One of the reasons to use a computer is to have it do the
work. I do have power options configured to spin down the HDDs
(obviously doesn't apply to the SSD). The monitor also powers down but
the CPU and rest of the computer is full on, so it is immediately
available for whenever I choose to use it.

My computer is asleep 80% of the time. Takes an additional few seconds
to wake it. Yes it has thermal cycles. Yes the hard drive spins down a
lot.

Users just don't understand about surge current on start or thermal wear
from heating and cooling repeatedly. For example, ever hear someone
recommend to reseat the cables, memory, or some other component? They
only "walk out" of their connections due to thermal expansion and
contraction.