I just concluded a decade long experiment. I had a USB flash drive in a jar buried in my back yard since 2015. I dug it up, plugged it in and it suffered no data loss after 11 years idle underground.
It's a usless experiment but everyone needs hobbies.
@Aaron_DeVries
While doing environmental testing on a helicopter payload, I learned that NVMe drives (and perhaps all SSDs) can write data at high ambient temperatures >40 C, but the data is less permanent than if you write data at normal ambient temperatures.
So on hot days we had to hurry and copy our 4TB drives after the flight because the data had a half-life of a dozen hours or so.
That phenomenon is already documented, but I don't think it's widely known.
@Aaron_DeVries @swope A lot of physical processes are thermally activated. At higher temperatures the barrier between two (meta)stable states is often easier to cross, which is why things can go wrong at higher temperature.
If it is, as I read it, bits _written_ at higher temperature having a shorter lifetime even at lower temperature after, I can imagine that the state written at higher temp isn’t as far down into the local energy minimum.
Yes, I imagine that something like that is going on. And the manufacturer is honest about the 35C max operating temperature.
For the user, they write a bunch of data and immediately verify the data are good -- not looking closely at the temperature log. But hours/days later their files become corrupted.
It's something to be aware of, and a curious physics question.
So, I guess we should store our testaments on an SSD while it is in liquid nitrogen
@sudo200
Lol. Ymmv
@swope @Aaron_DeVries I once saw a datasheet for an Intel SSD, which basically said it likes being run hot and stored cold, and that combination gave the longest data retention lifetime. Hot/hot, not so much.
If I remember correctly, that's because higher temperatures make it easier for electrons to escape from the floating gates. Which is bad for data retention, but as long as the drive is powered on the firmware can handle that by re-writing the data occasionally. But it also means deliberate erasures are easier and cause less physical damage, so the SSD degrades slower if it's run hot. And after enough use, that matters more than the effect of temperature on the durability of any individual write.
This is interesting.
Sadly, we didn't get the internal NVMe temperatures during our environmental testing.
I think the drives' spec said 35°C ambient for the max operating temperature, and we would typically see 45°C internal from `nvme smart-log /dev/nvme0` (I might not be remembering the command correctly).
I kinda supposed that there was an ideal temperature for writing that the drives try to maintain, but when ambient is too high they can't. Then what gets written is on the edge of being readable, and normal temperature storage losses degrade that in a few hours.
Lots of hand waving on my part. We just made a procedure and carried on. But it would be pretty insidious to look at your data right after recording and it looks good, but it's gone the next week.
@swope @Aaron_DeVries it’s inherent to all SSDs, to varying degrees, because it increases leakage from the cells — from memory offline storage at temperature is actually worse than online behaviour.
you can certainly design SSDs to be better at high temperature, mostly by doing the normal things to decrease leakage, like single layer cells, decreasing density of the cells, increasing the number of error correction bits, etc. you can also just mitigate the online case in firmware, by increasing cell refresh frequency.
such SSDs are manufactured, primarily for industrial, aviation, automobile, and similar applications, and especially targeting such markets you’ll often find data sheets giving a lot more detail about retention at temperature.
@job
No, I think it's more like using hot glue on plastic. While it's hot the glue sticks to the plastic, but as it cools your whole crafting project falls apart.
(Just a metaphor, not the physics)
I think that's intuitive for storage, but the hot-at-write-time weakness may be a little different.
I'm imagining an array of cups and the drive controller is pouring water in them as the write operation. When the system is hot, the cups are jiggling and aiming the carafe is shaky. Not a lot of water gets in the intended cups, and some spills into the wrong cups. Enough goes in for the theshold of the immediate validity check, though.
Then even at normal temperature, slow evaporation over days means the amount of water in those cups drops below threshold in enough cups to break ECC margins -- corrupting the files.
I don't have a deep enough understanding of the physics to tell you if that analogy has much validity. But this is my hunch.
@andi
Sorry I don't have the data available on the case or internal temperatures.
I agree that it was probably quite high.
"@Aaron_DeVries
Mientras realizaba pruebas ambientales en la carga útil de un helicóptero, descubrí que las unidades NVMe (y tal vez todos los SSD) pueden escribir datos a temperaturas ambiente elevadas (>40 °C), pero que dichos datos resultan menos permanentes que si se escribieran a temperaturas ambiente normales.
Por ello, en los días calurosos teníamos que darnos prisa en copiar el contenido de nuestras unidades de 4 TB tras el vuelo, ya que los datos tenían una vida media de unas doce horas, aproximadamente.
Este fenómeno ya está documentado, aunque no creo que sea ampliamente conocido."
That payload was writing continuously about as fast as the drives could go. Not a typical use case I think.
@swope @Aaron_DeVries Hard disks are similarly thrilling for flight. There is a maximum altitude (3048m). Over that the risk of the disk head crashing into the platter is too great.
That's why the datacentre for the Mauna Kea obervatories (9750m) is part way up the mountain.
Aaron DeVries
Leo@ALLES
swope
Jos Dingjan
clang-enby
Rachel Barker
Sam Sneddon 🏳️⚧️
EveyPN
Job Snijders
~n
AndiS 🌞🍷🇪🇺
lord pthenq1
Val Packett 🧉
the vessel of morganna
Glen T, heated, not stirred