[Novalug] SSD on sale

Bryan J Smith b.j.smith@ieee.org
Mon Jun 29 13:59:47 EDT 2015


On Mon, Jun 29, 2015 at 1:34 PM, Joshua Ellsworth via Novalug
<novalug@firemountain.net> wrote:
> Given the audience of this list, I think that taking a look at this article
> might be a good thing before jumping in on samsung SSDs:
> https://blog.algolia.com/when-solid-state-drives-are-not-that-solid/
> summary: Some Samsung SSDs have a bad trim implementation on
> linux that can cause data corruption or loss

Yawn ... old news.

Lesson:  _Never_ enable TRIM without testing it, with a decent
capacity.  You'll quickly know when it's screwing up.  ;)

I ran into this with a Mushkin 240GB mSATA device back in late 2012,
and had to disable TRIM.  Luckily I tested it first.  ;)


Bigger Lesson:  We (the Linux world) "lost control" of discard/reclaim
of blocks once NAND devices became commodity for Windows systems.

I.e., before the "consumer" adoption, we used to have to manage the
raw NAND (or NOR) device directly, and run a filesystem designed to do
such -- e.g., the JFFS* series.  That allowed the Linux kernel to have
far more direct and useful control of mapping, reuse, etc...  It
worked much better.

But since 90% of the market is cheap, NT-based Windows systems, and NT
cannot have a read-only boot (only CE did, even embedded NT uses an
overlay instead), we've been stuck with this "boundary" where we are
at the mercy of the commodity NAND device's firmware.

NAND is best for read-only applications.  Don't tempt fate and rely on
it for your business alone.  In fact, this is where dm-cache and other
things are really nice, using a smaller, cheaper NAND block device as
a read cache while it's 100% backed by platters in RAID.


BTW, Samsung has had some whoppers over its lifetime, especially early
models.  There's a reason I bought the 3rd gen 840 EVO, which were on
sale because the new 850 EVO was hitting, and I don't trust 1st gen
Samsung products.  ;)

But even Seagate, with this SSHD (platter-backed, caching NAND) has as
well, but at least in the case of the SSHD, the platter is the
"ultimate authority" of the bits.  I've never had any data loss, but
I've had a mis-read and a couple of crashes due to the NAND cache
being incorrect.

Lastly ... I do _not_ recommend RAID with NAND at this time.  Long story.

E.g., in my personal 15.6" notebook, I have a 1TB Samsung 840 EVO
1.2x2" mSATA (1.2x2") card, JBOD, and a pair of 2TB Samsung 5400rpm
3-platter 2.5" SATA (4.5x6") drives in RAID-1.  I backup key drives of
the former to the latter, and then both to an external, RAID-1 USB 3.0
bay with another pair of 2TB Samsung 5400rpm 3-platter 2.5" SATA
drives.

-- bjs



More information about the Novalug mailing list