Question Backing up data to SSD

[Mawla]

I have a SATA-USB adapter and am thinking of using some SATA SSDs as one method of backing up my data. I understand that SSDs can lose data over time from cell leakage so that they have to be "recharged" now and then.

So, is it enough to plug in the drive via the adapter every few months? And for how long each time - a few seconds, a minute or longer?

I'm aware that there can be no absolute guarantee of data safety with any storage medium. My question is about how the technology behaves.

 

[evermorex76]

Are you only going to make backups every few months? Or are you going to have a large number of drives that you rotate through over the course of months? Most people only really need to have a week's worth for home users, and just keep the drive plugged in all the time to do daily incrementals/differentials.

They only lose data if they're left unpowered for a really long time. If you're cycling through them, it won't be an issue. You really have to actually rewrite the cells to refresh them and avoid bit rot due to charge leakage. The wear levelling algorithms do some of that during normal usage, moving data around to make sure the cells are evenly used, but you can't be sure every bit of data was rewritten at some point, and I think just plugging a drive in for some time isn't going to be enough. (After all, the drive controller would also need to track how long it has been since each cell was last written to in order to know if it needs to be refreshed, otherwise it might be adding more writes than necessary and shortening the life even further.) Copy all the data from one drive to another, do a verification pass, then secure erase the original (or just delete everything and let TRIM/garbage collection have time to run). Or move it off of the drive temporarily, then move it back, but of course you're adding writes to whatever temporary location you use and you need all that space to do it.

If you're going to have 3 months' worth of backups and rotate drives, then you're not likely to lose any data to leakage on any during that time. If you're ARCHIVING them or having a really long backup rotation period and going to have them left unused for a year or more, then yeah, you might want to have an extra few so you can cycle through doing the transfer process, moving the "oldest" up in the queue each time.

With TLC and even worse QLC, the amount of time that you can let them sit can be much shorter than MLC or SLC, but a few months ought to be fine. Supposedly JEDEC requires them to last a full year unpowered before losing data, but that's a new drive, assuming it adheres to standards, and many have been found to last far longer. As they get used, the time shortens, but that requires a considerable amount of writes.

 

[Mawla]

Are you only going to make backups every few months? Or are you going to have a large number of drives that you rotate through over the course of months? Most people only really need to have a week's worth for home users, and just keep the drive plugged in all the time to do daily incrementals/differentials.

Let's keep it simple and think in terms of archiving data that will seldom need to be accessed or altered.

You really have to actually rewrite the cells to refresh them and avoid bit rot due to charge leakage. The wear levelling algorithms do some of that during normal usage, moving data around to make sure the cells are evenly used, but you can't be sure every bit of data was rewritten at some point, and I think just plugging a drive in for some time isn't going to be enough.

OK, let's say I have a drive installed in a computer that's in regular use but this particular drive is never read or written to after initially backing up some data on it. Are you saying that there's a good chance of losing data anyway due to bit rot? This is not about drive failure but about the natural process going on inside the drive.

 

[evermorex76]

OK, let's say I have a drive installed in a computer that's in regular use but this particular drive is never read or written to after initially backing up some data on it. Are you saying that there's a good chance of losing data anyway due to bit rot? This is not about drive failure but about the natural process going on inside the drive.

I expect it would be a possibility, if there is simply no activity at all, as there would be no blocks getting written or erased. (Windows though might just write data to a drive sometimes for no particular user reason so maybe that would help.) I think some drives at least do also have leakage mitigation in the controller which might actively do that sort of thing in the background, specifically added after the raft of issues when this was first found to be an issue.

I think 6 months would be fine, even a year or two. Consider your normal OS drive. How many gigs of data just sit there, having been written during installation and never needing to be changed? Even just the log files that aren't modified would be taking up blocks like that. We don't tend to worry about all those, and updates don't modify every OS file.

If you're really trying to archive, you need to be looking at tape, or maybe optical media if it's not REALLY long-term. Tape is better for mass archiving of course so you don't have to swap discs constantly. SSDs are NOT considered reliable for archiving, and even magnetic hard drives have turned out to have limited lifespans when they're just left unpowered for decades.

What is your other backup system? Is this the "take a copy once a month and send it offsite to be kept for a year or 5 years" part and you've got daily backups or hourly backups that will be kept locally and rotated more often, with a shorter retention period?