For tape, though, assuming the data is written linearly in raw TAR format, it should just be a matter of writing an amount of parity to places on the tape, and then taping it back together. I did this a while back using btrfs subvolumes and snapraid, and it can also be done with something like parchive. It’s just a pain in the ass to check and repair.
One of the great things about tape is, the data storage part isn’t really complicated and can be repaired.
I’m not sure how relevant it is to LTO though, because it’s such a locked down and unused format that there’s very little information on it. It could be that the drive will reject damaged tape entirely, but that seems really silly to me.
So, I think strictly speaking, the best for long-term data storage is probably custom-built open reel tape machines, or maybe LTO if tape repair is easy enough, but the best for a regular user is probably mechanical harddrives.
The mdisc bluray, at least last I checked were just really good bluray.
But the mdisc DVD are supposedly a litho something or other where the bits are physically altered ceramic?
I have burned the really important data to those, in addition to my normal 3-2-1. Just some documents and family photos that would suck super hard to lose. (Also backed up to cloud, but I have a hard time trusting cloud storage).
Some tidbits on this topic I’ve run into while I was doing some research a while back:
There is an SSD by Verbatim that is setup as a WORM (Write Once, Read Many) device that has a rated warranty of 10 years. I ran into the article about it within the last 2-ish years.
*Edit - found article and updated info
According to the documentation for the Samsung DCT 983 SSD’s, the data retention is rated for 2 weeks (I looked for and prolly missed exactly where that is stated when I looked for it again.)
There’s documentation stating that that as program / erase cycles are used, the cells are more easily prone to losing charge and the amount of time they are able to hang onto that charge is reduced.
One of the most abused areas of an SSD (if the workload is mostly small writes - eg random 4k) is the mapping table. It is never remapped to another region and is always the same location (unless this has changed recently - doubtful though).
This is true and this is why you want a drive with DRAM. What happens in those is that the mapping table is put in the DRAM and will only be written to the drive periodically and if needed. It really saves on the SLC writes, and as a bonus the drive is much better at sustained writes, too.
That said, DRAM is a workaround, the first sector of the drive should store a pointer to the mapping table not the table itself. That way the mapping table should be movable to a fresh-ish sector.
Reading through the product page it looks like the operation of refreshing the weak NAND cells is performed by the special software on the host PC itself as opposed to the embedded flash controller. Also it explicitly states its using MLC NAND.
Great topic. It’s disappointing how few SSDs give any data retention specs. It makes it hard as a regular consumer to know what to use for cold backups.
The WORM Vertbatiim SSD is very interesting, but Windows only writing is a big hassle for me.
I would like to think external SSDs like the Samsung’s T7 series would be designed to sit unpowered for long periods of time. But I can’t find any indication that it actually is
I guess mdisc will continue to fill a niche for me, we may not know if they really last 1000 years, but at least the manufacturer is willing to make some claim.
Anecdotally, I’ve had a 64MB flash drive from the early 2000s unpowered for over a decade without losing any data.
But, it’s 64MB, from 2001, and may not even use NAND flash, much less multi-bit-per-cell NAND flash.
Really, though, the solution I think is still harddrives for now.
Also anecdotally, I have yet to have SSD failure in any consumer gear since 2013.
I have a Samsung evo 850 still in use in my main desktop, 9 year old MacBook Pro, an 8 year old (I think) OCZ drive, etc.
Whilst I suspect it is heavily workload dependent, I think that since OCZ went broke and early SSD controller issues got sorted, NAND is much more reliable in my experience. I’ve gone through 2-3 sets of spinning disks in my NAS in that time frame due to drive failures/replacement cycles and the NAS sees no where near as much read/write as my primary devices.
Yes, in theory there are limited read/write cycles, but in reality for most use they’re more than you’d likely get out of a hard drive anyway. Partly because hard drives aren’t even fast enough to get there
I have two dead SSDs, one 2022 made Kioxia 500gig SATA and one ancient Samsung 120Gig (which served for 12 years before gracefully switching to read-only).
Switching to read-only is a very civilised way of behaving when a SSD approaches end-of-life. Nice one, Samsung.
I’ve had many — and exclusively — SSDs since 2010 and none of them have failed so far. Some (especially a 64GB Kingston) are getting noticeably slower, though… so the writing is on the wall. That said, can’t complain about drives that last 13 years.
A couple of SSD failures over here. Let’s see if I can remember them all…
Two ADATA SATA SSDs. Both failed and all data was lost. One (if not both) would only show up as a 1-ish MB device. This was pre 2017.
The thought of this kind of failure terrifies me tbh.
Seen a Kingston SATA SSD fail in the same manner. Pre 2017.
A SATA OCZ SSD failed. When plugged in at boot, would lock up the UEFI BIOS. When plugged into a running windows machine, explorer would lock up.
Linux reported it was read only and I managed to clone to another device.
This happened out of the blue and was still a new SSD well within warranty. Pre 2017.
RMA’d and received a refurbished unit that had a peculiar issue. Windows and Linux installs would eat dirt after a while. When HDD Sentinel did a surface scan, it reported a long list of bad blocks. I did a secure erase on it many years later and all of the bad blocks disappeared. It’s been running flawless ever since.
A friend of mine purchased a brand new from Newegg, a 2TB Samsung 990 Pro NVMe SSD that would randomly have io errors under windows and Linux. OS’s would eventually become corrupt and eat dirt. FW was up to date, etc. 2023.
oh yeah just thought i should mention. i have way more SSDs than mentioned. i just listed the ones older than 5-6 years, which i consider to be more than acceptable life.
A little over 10 years ago when MLC drives were becoming somewhat affordable we moved to them here at work. I had heard of the degradation issues, as had many other people at the time. Someone was nice enough to write a little utility that can be run and scheduled in Windows that would just read the data on a block and then re-write it on whatever schedule you wanted. I had it do this twice a year. That wasnt going to add any significant writes to the system to cause drive lifetime to go down, and it was nice to refresh the data. I wish I still had that little program. I dont even remember the name of it unfortunately.
edit:
Im curious as to whether XPoint has the same degradation problems? It uses the same storage principle, but is made of different materials. It has far higher endurance to the cells, but everything loses charge eventually. So I wonder if the XPoint cells degrade over a few years or if they leak so slowly that it takes 10-20+ to be a problem?
edit again:
I guess XPoint does not suffer degradation issues. It does not use the same storage principle like I thought. It does apply a voltage in a similar way to store a 1 or 0 in a cell like NAND does, but instead of actually storing a voltage level like NAND to determine a value XPoint instead changes the state of the material the cell is made out of more similar to a magnetic hard drive platter (change magnetic state) or like a CD or DVD media (optical laser to heat an area and change media state). To read the data, a very small voltage is applied and if a current passes it is a 1 and if no current passes it is a 0. The state of the material will hold “indefinitely” because the molecules are aligned that way already and don’t change unless they get very hot from a voltage being applied.
So case solved on that technology front. https://pcper.com/2017/06/how-3d-xpoint-phase-change-memory-works/
On the SSD side, I’ve had four Samsung 840 evo’s, one 850 pro and one sabrent rocket 4 fail on me, only the sabrent abruptly failed with no chance to get data off.
On the HDD side I’ve only had a 4TB WD SE and a 750GB white label Seagate fail me, both gracefully.
That sounds pretty useful. My current “solution” the NAND SSD slowdown issue is to boot up a live Windows environment that runs completely in RAM and have it run a surface test via hdsentinel on my SSD to rewrite all the data on it, about once a year is good enough for me.
I’ve mainly been running Ubuntu/Mint for the last 14 years on SSDs. Since unwanted ‘cruft’ accumulates whenever you upgrade an OS, I’ve been using a pattern of Upgrade → Clean Install → Upgrade → Clean Install… on a two-year cadence.
Every four years, therefore, boot drives get formatted, a fresh OS and applications are installed, and then home directories are restored from backup drives. Shortly thereafter, the backup drives are formatted, and fresh backups are made.
As a result, my active drive set is formatted — and data laid down fresh — every four years. I cannot remember ever having suffered from bit rot on a SSD.
(I’m not saying bit rot hasn’t happened, just that I haven’t ever had to restore a document or reinstall an application because it suddenly became corrupted or stopped working normally.)
In the 1990s I used to Upgrade → Upgrade → Upgrade → Upgrade… and my systems became slower and more unstable over time (as ‘cruft’ accumulated). Once I interleaved Clean Installs into the cadence, performance and stability issues went away.
tl;dr: Instead of always Upgrading your OS, perform Clean Installs every few years — it helps deal with bit rot as well as other issues.
from what i can find it still takes about a year (give or take a few months) for the gates to lose charge.
how do you recharge them… plug it into a working pc
just plugging it in will refresh the charge on all cells.
same with sd cards… 1 year give or take.
it depends on the size of gates, the wall thickness, as to how much charge they can hold.
but all these variables dont change much as the voltages to open or close the gates is proportional to the size.
which means regardless of node size, drain time will remain pretty consistent.
enterprise level storage may have a power reserve but you will pay silly money for it…
so back up to spinning rust if its for the long term.
