SSD as Lengthy Time period Storage Testing – HTWingNut Tech Weblog

Final Up to date on January 4, 2023 10:23 am

2022 September 10

What’s the greatest media for long run storage is a query that I learn steadily from on-line boards. At the beginning, so far as I’m conscious there isn’t a actual “set it and overlook it” long run storage. On this digital age, knowledge must be validated frequently and cycled to a brand new media each 5-10 years with a view to guarantee it’s protected and simply accessible.

Laborious drives have been round for about so long as the PC has existed, and proceed to be a strong cheap possibility for long term storage, though they’ve recognized longevity points. Whereas some exhausting drives might stand the take a look at of time, sustaining knowledge on them for 20 years or extra, that’s extra the exception than the norm. Now with SSD’s slowly taking on the duties of exhausting drives, it makes one marvel how good they could be at storing knowledge for prolonged durations.

SSD vs HDD Knowledge Storage

Whereas each SSD’s and Laborious Drives retailer knowledge as binary knowledge, how they do that is fairly totally different.

Laborious drives use one to many spinning platters the place a tiny electromagnetic learn/write head manipulates the floor of the exhausting drive to retailer knowledge of tiny magnetic particles to characterize 1’s or 0’s. There’s a tried and true historical past with exhausting drives and we’ve a good suggestion on longevity of exhausting drives and the way lengthy they’ll retailer knowledge. 5-7 years is typical earlier than the chance of degradation is elevated considerably.

SSD’s, however, retailer knowledge in “cells” and lure electrons in these cells utilizing floating gate transistors. The trapped electrons generate a voltage and that voltage is represented as a particular knowledge state. SLC SSD’s use only a single 1-bit change so 2^1 = 2 knowledge states: 1 or 0 per cell. MLC retailer 2-bits or, 2^2 = 4 knowledge states: 00, 01, 10, 11. TLC 3-bits or 2^3 = 8 knowledge states: 000, 001, 010, 100, 011, 110, 101, 111, and newest is QLC or 2^4 = 16 knowledge states: 0000, 0001, 0010, 0011, … don’t make me write all of them out, however you get the thought.

In a “contemporary” or “new” SSD the floating gate may be very strong and the chance of electron leakage is small. A effectively used SSD, nonetheless, these floating gates are topic to the legal guidelines of physics and put on down with use as electrons pour by way of them time and again. So in an idle powered off state, it could leak electrons, altering the voltage states of the cell. In SLC SSD’s it could actually tolerate a big voltage drop as a result of it solely must learn a 1 or a 0. However with one thing like QLC, when it has to learn/retailer a voltage that pertains to one in every of 16 knowledge states, a slight lack of electrons will lead to a voltage dip, which is able to lead to studying a special worth than was initially saved. This leads to knowledge corruption.

Apparently even plugging an SSD in a pc powered on received’t essentially maintain the electrons charged, however it could provoke a put on leveling routine or different algorithm that will maintain the information contemporary. However I’m trying to consider when you have been to maintain it unplugged from an influence supply.

Testing the Idea

The place that is all resulting in is a silly little take a look at I made a decision to run. I nabbed some extraordinarily low-cost Leven JS600 128GB TLC 2.5″ SATA SSD’s from Amazon at about $13 every and subjected them to 2 states, a set of two SSD’s “contemporary” and a set of two SSD’s “worn”.

Two SSD’s have been written to with over 280TB of knowledge, leading to roughly 2200 write/erase cycles. TLC NAND can usually face up to 1000-3000 writes earlier than they’ll begin to fail. So these SSD’s are effectively worn.

The opposite two SSD’s have been left new/contemporary with none writes apart from a single go of zeroes (Laborious Disk Sentinel / Home windows 10) and browse again to make sure nothing was manifestly mistaken with the SSD’s out of the field.

The “worn” SSD’s have been then wiped with a go of 0’s throughout the disk utilizing Laborious Disk Sentinel in Home windows 10, and browse again to ensure there have been no impending failures. CrystalDiskInfo ensured all disks have been in good well being

See Also

Speedy text-to-image technology on-device – Google Analysis Weblog

Every disk was formatted ExFAT and crammed with the very same set of random knowledge. MD5 hashes have been taken to make sure all of them matched the supply knowledge. This was accomplished on 2022 Sep 02. These SSD’s will likely be saved in a drawer in common local weather managed setting that I stay in and can solely be accessed with the next intervals:

Disk 1: “Worn” ~ 280TB written, 1 12 months: 2023SEP02, 3 12 months: 2025SEP02
Disk 2: “Worn” ~ 284TB written, 1 12 months: 2023SEP02, 3 12 months: 2025SEP02
Disk 3: “Contemporary” ~ 214GB Written, 2 12 months: 2024SEP02, 4 12 months: 2026SEP02
Disk 4: “Contemporary” ~ 214GB Written, 2 12 months: 2024SEP02, 4 12 months: 2026SEP02

They are going to be validated by verifying towards the MD5 hashes taken from the supply, utilizing hashdeep64. The intent will likely be to maintain it plugged in for as brief a time as potential with a view to decrease any SSD algorithm from refreshing the cells by shifting knowledge round.

Let’s see if I can correctly keep this process, and if we get any knowledge loss after 1, 2, 3, or 4 years’ time.