Hacker News Reader

Comment by izzydata | Hacker News Reader

What stops you from writing data to a spinning disk HDD and then unplugging it and putting it in a safe?

Mechanical failure [0]. As perbu found out, when moving parts are left to sit for years, things can fail. Lube can dry up, causing bearings to seize up, actuators to lose precision or get completely stuck [1], and the magnetic field degrades by 1% per year [2]. Of course, you could get lucky w/r/t mechanical issues, as many retro computer enthusiasts can attest, but it seems to involve a lot of luck, as it only takes tiny defects to cause problems over a long enough time-frame that leave half your drives as paperweights while the others keep on chugging.

Both of these issues can be largely mitigated by simply plugging the HDDs in every year or so and re-writing the data (maybe run badblocks while you're at it to ensure the drive is still in good health), but, as throw0101a mentioned, this takes time and effort.

[0]: https://serverfault.com/a/51893

[1]: https://www.partitionwizard.com/clone-disk/do-hard-drives-go...

[2]: https://superuser.com/a/312764

Is there a way to force a file system to recommit its data?

Seems like you could do it by not even mounting the filesystem, just read out raw sectors and write them back. Like a 3 line program. Maybe if you want to be fancy you compare the sectors to some checksums to make sure you're not reading corrupt data.

Any word on SSDs?

SSDs may actually worse than HDDs, depending on what NAND chips are in use. Although there are no moving parts to seize up, they do require an occasional power up because they represent data via voltage levels and we have not yet found a perfect insulator (meaning they will inevitably leak charge), and with QLC being all the rage these days (4 bits represented in each cell using one of 16 discrete voltage levels), when the voltage levels start to degrade, it could easily cause catastrophic bit rot.

How long a given SSD will last before losing data depends on the chips it is built with. I remember reading that Sandisk rates their flash storage as being able to survive a year without power (though I read that years ago and cannot find a source for it now), though [0] says 2-5 years, so technology has likely advanced since I last read about this. Either way, you should probably try and power up your SSDs at least once a year to keep the cells charged. This is less work than to keep an HDD healthy for cold storage, but it certainly isn't "set and forget" like tapes or optical disks.

Edit: I remember pulling my first SSD, a 60 GB Intel 330, from an old laptop and, despite having not been powered on for 3 years, it still booted. But that was MLC/DLC [1], which only has 4 voltage states, and I only saw that it booted, not that all my old data was intact.

[0]: https://www.easeus.com/resource/does-ssd-need-power.html

[1]: https://www.intel.com/content/dam/www/public/us/en/documents...

Hm. Makes me more sad about the state of zfs in macOS and windows - with native encryption and checksums it makes for a nice cross-platform fs for external drives...

> What stops you from writing data to a spinning disk HDD and then unplugging it and putting it in a safe?

Manual labour.

There are tape libraries with robots that can take a tape out of a slot, put it in a drive, and then put it back automatically without humans needing to do anything. If you want to move the tapes offsite, there is some manual work to move the tapes from the import/export slot to a container.

A large library like the StorageTek SL8500 can have up to 640 drives, 100'000 tape storage slots, and 2'880 import/export slots.

> Manual labour.

Big cloud providers have disk replacement robots, so it's not unfeasible for robots to perform the "unplug and store the HDD somewhere".

But another advantage tape has is durability.

> Big cloud providers have disk replacement robots […]

[citation needed]

https://twitter.com/alibaba_cloud/status/1326078390105825281

I have no idea if anything like this is actually in use, I have my doubts. Also, a system like this probably isn't terribly useful, a Cartesian gantry robot would be much faster and way more robust than an articulated arm strapped to a mobile robot platform.

Who actually has those?

Last time I saw one of those was back when AOL was a big shot ISP. They had to buy high power SGI systems to feed the things because nothing else had the bus bandwidth to support one. I was told that the full rack sized SGI system (Origin 2000) was the cheapest part of the entire rig.

> A large library like the StorageTek SL8500 can have up to 640 drives, 100'000 tape storage slots, and 2'880 import/export slots.

Wow the scale is just immense, that's pretty cool!

Typical reliable offline HDD storage is 2-3 years. For tape it's 20-30 years.

I've got a bunch of video on a harddrive that I put in storage 20 years ago. Tried to spin it up again and it just makes funny noises.

So, yeah, spinning rust might rust.

its possible to restore it in a data restoration shop, if you are interested

This is what I personally do. I have small needs when it comes to backups (< 1TB), so I have one backup that's just a regular HDD stored in a drawer, and another one that's on a Scaleway cold storage container.

The spinning motor might not start after some years of inactivity.

Bitrot.

Reply on news.ycombinator.com