[crazygringo]

This article mentions disposing of hard drives that have reached their five-year mark and are no longer under warranty.

Does anybody actually want hard drives this old? Isn't the whole point that the risk of failure and therefore data loss is too high by this point? Even if you're using them to store data redundantly, you're running the risk that when one drive fails, the backup will also encounter failure due to the stress of reading its entire contents at once in the attempt to create a new backup.

[AnthonyMouse]

> Even if you're using them to store data redundantly, you're running the risk that when one drive fails, the backup will also encounter failure due to the stress of reading its entire contents at once in the attempt to create a new backup.

There are RAID levels with N+2 redundancy, or more with some of the fancy stuff.

You're also getting stats from Backblaze, which have the hard drives in a server, constantly powered on, under significant load. A drive that was sitting on a shelf isn't going to be less reliable just because the manufacture date is three years ago and the warranty has expired. A ten year old drive can have the same number of power on hours as a one year old drive.

Moreover, sometimes the data isn't unique. If you need a drive to host a mirror for some Linux distro it's not like you're hosting the only copy in the world.

And if the data is critical, you need better than RAID regardless. What's your plan if a voltage spike takes out multiple drives in the same machine at once? Lightning doesn't care how old your drives are.

[Krisjohn]

I work somewhere that specialises in secondhand enterprise equipment and storage. The answer to your first question is "yes, but they really shouldn't". We're burning more hours than we can spare dealing with drive failures. Sure, we have RAID arrays capable of withstanding multiple failures, but we still find ourselves scrambling to replace drives faster than they're failing, even when they were tested before they were put in the spare pile.

Also, fuck Chia. It was supposed to be a low-power "proof of storage", but it's really "proof of prior work" and burns even more energy than proof of work since you need to constantly power the cryptographic calculations AND storage.

[justajot]

This statement on using more energy than proof of work is ... incorrect. How did you come to this conclusion?

[JohnFen]

> Does anybody actually want hard drives this old?

Sure, I would! I frequently use hard drives much older than this, and while I know there's an increased risk of failure, it has never happened to me -- so that risk appears to be quite tiny.

[crazygringo]

Take a look at how failure rates dramatically accelerate once you hit 5 years:

https://www.backblaze.com/blog/how-long-do-disk-drives-last/

The risk is anything but tiny.

"How long do drives last? It would appear a reasonable estimate of the median life expectancy is six years and nine months."

[JohnFen]

I'm not questioning those statistics, really. Nonetheless I'll continue to go with my own experience on this. I typically have hard drives in use for 7-10 years or so before replacing them for something with higher capacity. This has never caused any issue. And when I replace a hard drive, I keep the old one in storage as an extra backup. I've copied data off of 20+ year old drives from storage without a problem before (although that's a bit different than using them daily).

In any case, should a hard drive fail, it's not of great significance because of redundancy and backups. Also, spinning platter drives usually give plenty of warning of impending failures.

Used hard drives remain very attractive to me.

[EMCymatics]

It is wild how many failure rates coincide with warranty.

[toast0]

If you have a good plan to recover from failures and you regularly monitor for pre-failure indicators, that the warranty expired shouldn't be a reason to drop the drives.

A hard drive being in warranty or not doesn't indicate much about its likelyhood of working. There's a market for 5 year old hard drives that seem to be working, and at the same time, if you have budget, replacing your hard drives every 5 years will likely get you decent incremental capacity increases.

[pixl97]

> not doesn't indicate much about its likelyhood of working. T

Statistically I would say it does very much.

https://en.wikipedia.org/wiki/Bathtub_curve

Also drives can start building up bad sectors that you cannot write to, but may be able to read data from.

[toast0]

The bathtub curve is real, but it's hard to know where it is for a given model and production date. My personal experience is that hard drives often continue working past their warranty date, so I'd guess the other end of the bathtub is closer to ten years than five. I've run server fleets with a few thousand drives, and didn't find the other end of the curve because five year old drives are both out of warranty and relatively low capacity; we would retire systems with old drives because a new system would have much more capacity, rather than because the drives were old, but we still didn't run too many old drives.

> Also drives can start building up bad sectors that you cannot write to, but may be able to read data from.

Bad sectors are a pre-failure indicator. It's totally reasonable to stop using drives when they collect enough bad sectors. My threshold is 10 for drives you don't regularly monitor and can't easily replace, and 100 for drives with automated monitoring and simple replacement procedures.

I wasn't ever able to figure out reliable pre-failure indicators for ssds. In my experience they work nearly perfectly, until they disappear, never to respond to commands again. Thankfully, at a much lower rate of failure (per drive) than mechanical disks.

[pessimizer]

With 6-8 drives in RAID 6, that's a vanishingly small probability which you should have an offline/remote backup to make sure would not be a disaster anyway. If there were a super-cheap source of five year-old drives, I'd never use anything newer.