[kwlug-disc] OT: SSD disks?

[Chris Irwin]

On Tue, Nov 10, 2015 at 12:08 PM, B.S. <bs27975 at yahoo.ca> wrote:

> Bear in mind the SSD warnings though - swapping increases SSD wear,


I think worrying about swapping to an SSD isn't necessary now.

Tech Report did an SSD endurance experiment with a few popular SSDs until
they all failed. Spoiler: Don't buy


http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

I did some math when the topic came up on another site (so the math is two
weeks old). I've got an OCZ Vertex 3 drive (uses the same controller as the
Kingston in the techreport test).

I've owned the drive for 1144 days (including shipping). It has been
powered on for 427 days (or 37%) according to SMART.

* I use this as my primary drive with root, home and swap.
* I've got an 8GB maildir, which is also in git, and use the (default)
relatime so mutt can correctly scan the thousands of files.
* I have swap
* This has gone through upgrades and reinstalls from Fedora 17 -> 23, as
well as regular system upgrades, file system conversions (ext4-btrfs, etc),
etc.
* Other than enabling the fstrim service, and allowing discards through
dm-crypt, I have done nothing special.

According to tech report, I can expect read errors at about 700TB of
writes. According to SMART, I've written 4TB. Granted the wear levelling
will get better results from more free space (as you noted).

The earliest drive to fail (and by quite a margin) was the Samsung TLC
drive. It started reallocating sectors at 200GB. Even applying that
guesstimate to my (daily driver) laptop, I've got quite a lot of breathing
room.

Now that we have TRIM, and all the distros are correctly aligning their
partitions, there is almost zero special consideration required now.

Regardless of your confidence, you should still treat *every* drive as if
it was about to fail, and back up accordingly. This isn't just an SSD
thing. I've got a nice stack of traditional hard drives with significant
errors (and a lovely seagate 7200.11 doorstop)


> Let alone SSD friendly mount options such as noatime, nodiratime, discard.


I had always considered noatime to be HDD tweaks: Less dirty data to spin a
disk up for, and fewer seeks. Things that are not considerations on an SSD.


> Mind you, I believe I've seen test results that show SSDs are at least as
> reliable as spinning disks now, which is to say at least a 5 - 10 year
> lifespan. (Never mind I've have more than a few spinning disks that are
> rather older than that.)
>

Using the tech report figures, you could write the current Sandforce-based
drives for 170GB/day for almost ten years before rewrites occur. again,
actual mileage will likely vary a bit based on free space, data turnover
(for lack of a better term), etc.

Khalid mentioned that his failed drive was an OCZ Vertex. The older OCZ
drives had obscene failure rates (some products were over 40%). I had an
OCZ Agility fail without warning. There's a reason they went bankrupt.
Their current drives (after the Toshiba purchase) seem to be of acceptable
quality, however.

Apparently you can help swapping / memory footprint by regularly minimizing
> browsers such as Firefox. Apparently it turfs some cache at minimization,
> or something.


I haven't heard of this. Interestingly, minimizing a window in Windows
increases the chances that it will be swapped.


> Certainly on my laptop it is the browsers that significantly impact
> performance of everything else. Run firefox and chrome together, both with
> many tabs open, and things start grinding.
>

The only browser performance issue I've had recently was with my Thinkpad
W520. On battery power, it refused to clock faster than 800MHz, while
firefox was making performance calculations based on the full (roughly)
2.4GHz clock frequency. Until I found the workaround, I couldn't play any
media (youtube, etc) and page scrolling would stutter.

-- 
Chris Irwin
<chris at chrisirwin.ca>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://kwlug.org/pipermail/kwlug-disc_kwlug.org/attachments/20151110/4ab6831e/attachment.htm>