• SSDs, trim (was: Re: molasses, last few days?)

    From J. P. Gilliver@3:633/10 to All on Tuesday, November 17, 2026 22:12:54
    On 2026/11/17 21:10:59, Carlos E.R. wrote:
    []
    Deleting files in an SSD doesn't work immediately. The sectors have to
    be actually cleared which is done by a process in background. I think it runs inside the disk firmware. It may be called "trim".

    Ask the Google AI "what does "trim" do on an SSD disk?" and you get a
    good answer about it :-)

    When I restored my C: partition, the Macrium software asked if it should
    do something (presumably it did it to th whole SSD) involving trim. It
    said something like "even on OSs that know about trim", implying that
    what it (Macrium) was offering to do was better than what the OS does; I
    said yes anyway, and nothing's died yet :-)
    []
    Can't say, I'm not really an expert on SSDs :-)

    Me neither.
    --
    J. P. Gilliver. UMRA: 1960/<1985 MB++G()ALIS-Ch++(p)Ar++T+H+Sh0!:`)DNAf

    You need 10 wise men to get back a rock thrown in a lake by an idiot
    - Romanian proverb (tweeted by Ovidiu S @plount_os 2023-2-6)

    --- PyGate Linux v1.5.19
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Tuesday, November 17, 2026 20:11:55
    On Tue, 11/17/2026 5:12 PM, J. P. Gilliver wrote:
    On 2026/11/17 21:10:59, Carlos E.R. wrote:
    []
    Deleting files in an SSD doesn't work immediately. The sectors have to
    be actually cleared which is done by a process in background. I think it
    runs inside the disk firmware. It may be called "trim".

    Ask the Google AI "what does "trim" do on an SSD disk?" and you get a
    good answer about it :-)

    When I restored my C: partition, the Macrium software asked if it should
    do something (presumably it did it to th whole SSD) involving trim. It
    said something like "even on OSs that know about trim", implying that
    what it (Macrium) was offering to do was better than what the OS does; I
    said yes anyway, and nothing's died yet :-)
    []
    Can't say, I'm not really an expert on SSDs :-)

    Me neither.


    https://www.techpowerup.com/ssd-specs/crucial-bx500-480-gb.d959

    "Crucial guarantees an endurance rating of 120 TBW,
    a relatively low value compared to other SSDs."

    And that's a value quoted for a 480GB SSD, so you would multiply
    that by 2 to get a comparative number to someone elses 1TB drive.
    My Daily Driver is 600TBW for a 1TB drive, your family is 240 TBW
    at a 1TB drive.

    This means that every flash location can be written 240 times.
    The low number suggests a write-amplification problem as the
    drive "handles" the condition of the writes. Things are squeezed
    into blocks ("consolidated") after some random writes, and this
    can be done in the background when the drive is not busy. If you
    used a CrystalDiskMark 4K random write test, it can take all night
    for a drive to undo the mess that makes. Just as an example of how
    much work it is to do background maintenance on the older drives.

    You're at 95%, so only 5% of 240 writes have been done so far.
    Each location has been written 12 times. Plenty of life to go.

    And it is TLC flash, so it can get "mushy" and the error corrector
    will be working hard on reads. Most of my SSDs in the room are TLC,
    and some of them are really bad for mush. The Samsung by comparison
    seems relatively good. The read rate can drop from 500 down to 150
    on a drive fighting a mush problem on reads. This is why we hate
    the SATA drives that don't have good maintenance code (or, actual good
    TLC flash).

    I would not touch QLC flash with a barge pole. And the five bit
    version they've been working on for some time, it hasn't been
    deployed yet.

    There have also been a few extra-large drives put on the market,
    but the price is so high, they might as well not exist. That's
    what you call a paper launch, it takes a lot of paper to afford one.

    Paul

    --- PyGate Linux v1.5.19
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From J. P. Gilliver@3:633/10 to All on Wednesday, November 18, 2026 12:30:09
    On 2026/11/18 1:11:55, Paul wrote:
    On Tue, 11/17/2026 5:12 PM, J. P. Gilliver wrote:
    []
    When I restored my C: partition, the Macrium software asked if it should
    do something (presumably it did it to th whole SSD) involving trim. It
    said something like "even on OSs that know about trim", implying that
    what it (Macrium) was offering to do was better than what the OS does; I
    said yes anyway, and nothing's died yet :-)

    (Any idea on what that's about?)
    []
    This means that every flash location can be written 240 times.
    []
    You're at 95%, so only 5% of 240 writes have been done so far.
    Each location has been written 12 times. Plenty of life to go.

    The _symptoms_ of my molasses experience made me suspect the "drive" -
    some things happened quickly, some glacially, as if those that were in
    RAM were fine. But as explained it turned out to be nothing to do with
    it. This is the first computer I've had with an SSD as the main drive; I decided, since I had to have Windows 10, that I'd better go for an SSD
    because of what I consider the blazing laziness of Windows 10 (up to W7,
    I stayed with spinners), though with trepidation - though I must admit
    I've had no problems with it, even running it nearly full for some time.
    (I still image it monthly though [and backup my genealogy and
    Thunderbird weekly]!)

    I still feel happier using a spinner for backup (though admit the weekly
    ones are on USB sticks).
    --
    J. P. Gilliver. UMRA: 1960/<1985 MB++G()ALIS-Ch++(p)Ar++T+H+Sh0!:`)DNAf

    Puritanism: The haunting fear that someone, somewhere, may be happy.
    - H.L. Mencken, A Mencken Chrestomathy

    --- PyGate Linux v1.5.19
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)