11 comments

[ 608 ms ] story [ 98.9 ms ] thread
I wonder how long those drives can be powered off before they lose the data. And until they lose all functionality when the critical bookkeeping data disappears.
Have had enough consumer SSDs fail on me that I ended up building a NAS with mirrored enterprise ones...but 2nd hand ones. Figured between mirrored and enterprise that's an OK gamble.

Still to be seen how that works out in long run but so far so good.

> Overall, I haven’t seen many issues with the drives, and when I did, it was a Linux kernel issue.

Reading the linked post, it's not a Linux kernel issue. Rather, the Linux kernel was forced to disable queued TRIM and maybe even NCQ for these drives, due to issues in the drives.

I wonder whats the best SATA SSD (M.2 2280) one could get now?

I have an old Asus with a M.2 2280 slot that only takes SATA III.

I recall 840 EVO M.2 (if my memory serves me right) is the current drive but looking for a new replacement seems not to be straightforward as most SATA is 2.5 in. Or if its the correct M.2 2280, its for NVMe.

Since it’s kind of related, here’s my anecdote/data point on the bit rot topic: I did a 'btrfs scrub' (checksum) on my two 8 TB Samsung 870 QVO drives. One of them has been always on (10k hours), while the other hasn’t been powered on a single time in 9 months and once in 16 months.

No issues were found on either of them.

> The reported SSD lifetime is reported to be around 94%, with over 170+ TB of data written

Glad for the guy, but here are a bit different view on the same QVO series:

    Device Model:     Samsung SSD 870 QVO 1TB
    User Capacity:    1,000,204,886,016 bytes [1.00 TB]
   
    == /dev/sda
      9 Power_On_Hours          0x0032   091   091   000    Old_age   Always       -       40779
    177 Wear_Leveling_Count     0x0013   059   059   000    Pre-fail  Always       -       406
    241 Total_LBAs_Written      0x0032   099   099   000    Old_age   Always       -       354606366027
    == /dev/sdb
      9 Power_On_Hours          0x0032   091   091   000    Old_age   Always       -       40779
    177 Wear_Leveling_Count     0x0013   060   060   000    Pre-fail  Always       -       402
    241 Total_LBAs_Written      0x0032   099   099   000    Old_age   Always       -       354366033251
    == /dev/sdc
      9 Power_On_Hours          0x0032   091   091   000    Old_age   Always       -       40779
    177 Wear_Leveling_Count     0x0013   059   059   000    Pre-fail  Always       -       409
    241 Total_LBAs_Written      0x0032   099   099   000    Old_age   Always       -       352861545042
    == /dev/sdd
      9 Power_On_Hours          0x0032   091   091   000    Old_age   Always       -       40778
    177 Wear_Leveling_Count     0x0013   060   060   000    Pre-fail  Always       -       403
    241 Total_LBAs_Written      0x0032   099   099   000    Old_age   Always       -       354937764042
    == /dev/sde
      9 Power_On_Hours          0x0032   091   091   000    Old_age   Always       -       40779
    177 Wear_Leveling_Count     0x0013   059   059   000    Pre-fail  Always       -       408
    241 Total_LBAs_Written      0x0032   099   099   000    Old_age   Always       -       353743891717
NB you need to look at the first decimal number in 177 Wear_Leveling_Count to get the 'remaining endurance percent' value, ie 59 and 60 here

While overall it's not that bad, losing only 40% after 4.5 years - it means what in another 3-4 years it would be down to 20% if the usage pattern wouldn't change and the system wouldn't hit the write amplification. Sure, someone had that "brilliant" idea ~5 years ago to use a desktop grade QLC flash as a ZFS storage for PVE...

The critical failure profile is when you fill them up almost full and then hit them with a bunch of writes. If you can avoid that you’re good for years.
My 3 of the first SATA SSDs are still in use from over a decade ago... I first had them in my home server as OS and Cache drives respectively, they later went into desktop use for a couple years when the server crashed and I replaced it with a mini pc that was smaller, faster and quieter. Then they eventually wound up in a few desk-pi cases with RPi 4 8GB units, and I handed them off to a friend a couple months ago. They're all still working and only 1 error between the 3 of them. IIRC, all 240gb crucial drives from early 2010's.

I've never had any spinning drives come close to that level of reliability. I've only actually had one SSD or NVME fail, and that's the first gen Intel drive in my desktop that had a firmware bug and one day showed up as an 8mb empty drive. It was a 64gb unit and I was so impressed by the speed, but tired of symlinking directories to the HDD for storage needs I just bumped to 240gb+ models and never looked back.

Currently using a Corsair MP700 Pro drive (gen 5 nvme) in my desktop. Couldn't be happier... Rust and JS projects build crazy fast.

I have used Samsung PM897 1.92TB for a year and it's okish... it is slow on the indexing when folders become saturated and as the drive fills up it becomes slower.

Solution is to remove some files... and pray it lasts half of a 64GB Intel X25-E!

It should last 30x shorter because it is 30x larger?

Or is this game only about saturation rate?

What size of apartment does an hdd noise become an issue in?
It depends on background noise, not sure how objectively it can be measured.

My apartment is super quiet, you hardly hear anything from the outside, so I can hear a HDD in the living room during the silent parts of movies. In some relative’s house, however, you only notice how loud is the background noise when the power goes off. No wonder why I always have a headache when I go there.