[sf-lug] "Stick PC" and other SFF computers

[Rick Moen]

Quoting Ken Shaffer (kenshaffer80 at gmail.com):

> Yes, it wouldn't be the pictures that wear out the flash, it would be
> the directories getting 100s of k entries.  Since I'm not reading the
> pics much, going with the default mount options for ext4 work (not
> even using the noatime option). 

A sane usage model, and I'm glad the flash storage is holding up.  I'm
actually impressed that you're having this good fortune with a microSD
card, since, unlike SSDs, they have no onboard circuitry to do
"wear-levelling" and that sort of thing.

When NAND flash storage was really new, I remember a lot of us were
quite wary of _all_ kinds of flash, as to long-term reliability, and did
things like having most-active trees on RAMdisks that were occasionally
written back to main (flash-based) storage.  The biggest worry, at the
time, was that flash-based storage was then notorious for sudden
catastrophic failure with no warning symptoms, whereas a hard drive
that's soon to fail often gives off noticeable signs of trouble, giving
you time for immediate backups and retiring the thing.

Since that time, manufacturers have found ways to make flash generally
better, and also experienced uses have gotten more comfortable with the
devices.  And also, it needs to be said, too, it sank in around the same
time just how badly essentially all hard driver manufacturers literally
engineer their drives to lie about their health.

E.g., you're probably familiar with SMART, Self-Monitoring, Analysis,
and Reporting Technology.  For decades, every HD has had a
self-monitoring layer that can be queried using SMART software,
such as the daemon smartd(8) from the smartmontools
(https://wiki.archlinux.org/title/S.M.A.R.T.).  Well, guess what?  HD
manufacturers were fully aware that customers were checking on drive
metrics using SMART clients, so they started making the lower layers of
their drive electronics tell the SMART layers "Hey, nothing going wrong,
here.  Everything's great.  No patterns of progressive sector failures
at all.  Stop worrying!"  And then, of course, maybe the drive
catastrophically fails, because of the patterns of progressive sector
failures that the SMART layer never heard about.

The result is:  Maybe you can trust what smartd is telling you.  Or
mayybe not.  Great, huh?