David AmadorFeb 6

TIL that SSDs can lose data if left unplugged for long periods of time (only required to hold data up to 1 year), unlike HDDs which as long as the material holds it can take years.

Edit: added link: https://www.slashgear.com/1893447/dont-leave-your-old-ssd-unplugged/

Show threadMarcus MüllerFeb 6
@djlink that is a very poor-quality source; modern SSDs indeed hold data for years, and powering them also doesn't increase data retention; they're not in any technical sense related to static (which needs constant power, very little) or dynamic RAM (which needs refresh cycles every few milliseconds).
You can be pretty certain that a not end-of-write-life SSD will retain data for years to decades. If you care, some SSDs actually specify more than just a overall MTBF (often in the 10⁶ h)
Show threadRay McCarthyFeb 6

@funkylab @djlink
Not powered off they don't.
An HDD can wear out with use but 25 years is easy for storage in a drawer or box in the attic. Floppy storage is far trickier.
Tape needs carefully stored.
Pressed DVDs* and especially pressed CDs are OK, but "written" ones can fade in daylight.

[* Assuming no manufacturing defects]

Show threadMarcus MüllerFeb 6

@raymaccarthy @djlink Ray, I'm sorry, but do you actually understand how flash memory works? powering on the SSD does exactly *nothing* to the cells until you at least read them (in which case you get a slight read wear on the cell and its neighbors), and you won't increase the charge levels inside a cell unless you erase and rewrite it, which does more damage, so the speed of charge leaking is higher than if you've just let the data alone.

(I mean, you're an EE – so model your gate capacitor!

Show threadMarcus Müller
@raymaccarthy @djlink … the charge it holds is your bits (mutliple, because on most SSDs these days you get more than two states); the only way that loses data is by tunneling of charge through the dielectric, which follows a shot noise model. Information-theoretically, we call this a "Z-channel", because you can only get from higher to lower states, never the other way around.
Now, if left alone, a couple of these bits will actually flip – that's why there's extensive forward error correction –
Feb 6 at 9:15pmPhanpy
Show threadMarcus MüllerFeb 6
@raymaccarthy @djlink but as long as the number of these high->low state degradations is small enough, that's correctable. Flash memory is already, without long-term storage effects, a lossy medium, which you have to design your error correction for!
So, when an SSD manufacturer designs that error correction (namely, which code to use, and thus how many bits per information word to add as redundancy), they have to design it in such a way that at some erase-write cycle reliability they want to
Show threadMarcus MüllerFeb 6
@raymaccarthy @djlink sell to the customer. But that same redundancy that helps when cells' dielectric layers degrade due to repeated high-voltage "zapping" (right, you apply a high |E| to the cell to implant charge in flash memory!) and doesn't hold charge as well also helps with long-term storage. Just that the effect of "time and temperature", as you can imagine, is a lot smaller than the effect of "make that dielectric experience what would be called a breakdown if it was macroscopic"!
Hence
Show threadMarcus MüllerFeb 6
@raymaccarthy @djlink I'm really not sure where the idea that a powered SSD would be more reliable than an unpowered one – that could only be true if it would be re-writing itself in the background, which would, counter to the intent, make it wear out faster, unless the SSD is essentially unused and the re-writing was free to use arbitrary much rarely or never used pages to copy the data to. But even that would be very undesirable – who wants an SSD with a standby power usage as if written to?)
Show threadRay McCarthyFeb 6
@funkylab @djlink
No, the main point is that an unpowered SSD isn't reliable compared even to floppies (though they are sensitive to storage conditions). I've had difficulty with 20 to 40 year floppies due to poor storage.
I transferred the MFM HDD contents to IDE HDD over 20 years ago. They are gone.
Show threadMarcus MüllerFeb 6
@raymaccarthy @djlink you must have had fantastic floppies, mine failed all the time; and I mean, the bit flip probability of 1.44 MB floppies, that's several orders of magnitude worse than that of an SSD even on undisturbed readout! I meanv how many bit flips do you think you'll see when you read 1000 freshly written to floppies? Certainly more than one! you can read that much data from an SSD in less than a second - and it will with probabilities of much better than 1 in 10⁶ have no bit error.
Show threadThe Penguin of EvilFeb 7
@funkylab @raymaccarthy @djlink Modern SSDs do patrol scrubbing so they do indeed rewrite bad information or marginal data when needed.
Spinny rust has btw done the same thing for many years too, some spinny rust is even smart enough to relocate data behind your back so you don't even find out about the bit of the disk that's getting dodgy for some reason.
Show threadRay McCarthyFeb 7
@etchedpixels @funkylab @djlink
SMART reporting on HDDs should reveal lots. Most Linux distros have a tool already installed to read SMART data and test HDDs.
Show threadAnthropyFeb 7

@funkylab Honestly you come across as condescending with your posts. Please don't.

- Trimming, voltage drift scrubbing, and wear leveling will cause data to constantly get re-written.
- While Single Level Cell NAND can hold your data for a decade, QLC can lose data within 1-2 years when left unpowered.
- Controllers are aware of this and can counter this if they're powered.

Example sources:
- https://www.xda-developers.com/your-unpowered-ssd-is-slowly-losing-your-data/
- https://www.seagate.com/files/staticfiles/docs/pdf/whitepaper/tp618-ssd-tech-paper-us.pdf

SSDs are leaky capacitors.

@raymaccarthy @djlink

The unpowered SSDs in your drawer are slowly losing your data

SSDs aren't ideal for long-term data storage

XDA