A hard drive sitting on a shelf is a big question mark. Will it spin-up? Who knows?!
So you deal with this by having a backup. Now you have 2x the cost.
But, unfortunately, you could have two hard drives not spin up. You consider an additional backup, but at this point you're at 3x the cost, so you start to reconsider letting hard drives sit on shelves.
Now you're in RAID country. RAID is not a backup, but by shifting your focus to availability you now have an approach to dealing with the "will it spin-up?" issue by proactively dealing with the problem via monitoring and rebuilds when necessary. You keep several replacement HDDs on hand for this inevitability.
Now, since you know RAID is not a backup, you put a backup system in place. This is additional cost; even more if you go 3-2-1.
So now you've got hard drives, storage arrays, HBAs, backups, and the power all of this consumes. We're starting to get out of cheap territory.
I suppose what I'm trying to say is: If you have a lot of data, hard drives are only cheap if you don't care availability, integrity, and longevity.
I hate comments like this. This is Hacker News, we are highly tech educated and wealthy on average. I specced a 71 TB usable array that can survive losing 3 of 8 drives.
$2000. (shucked drives, but this includes the server and ECC)
If the data is invaluable, throw down a couple of these at homes of friends and family.
I have 10 x 20 TiB volumes, software RAID6, across multiple storage arrays. I wrote my comment from experience.
I still don't consider it "cheap" even though I make a good chunk of change like many on this website, but I think this is because I think in terms of what my friends & family could afford.
Brings up an interesting topic: if I wanted to preserve a movie for 100,000 or even 1,000,000 years, how would I do it? What physical media would last that long, and could actually be played back correctly and accurately by some archaeologist in the year 1,002,025?
Plated of quartz, etched internally via focused femtosecond pulsed-laser to create tiny voids or just changes in crystal lattice that encode bits as changes in the polarization of light passing back through the glass.
It's been studied in various ways for a while now.
So you deal with this by having a backup. Now you have 2x the cost.
But, unfortunately, you could have two hard drives not spin up. You consider an additional backup, but at this point you're at 3x the cost, so you start to reconsider letting hard drives sit on shelves.
Now you're in RAID country. RAID is not a backup, but by shifting your focus to availability you now have an approach to dealing with the "will it spin-up?" issue by proactively dealing with the problem via monitoring and rebuilds when necessary. You keep several replacement HDDs on hand for this inevitability.
Now, since you know RAID is not a backup, you put a backup system in place. This is additional cost; even more if you go 3-2-1.
So now you've got hard drives, storage arrays, HBAs, backups, and the power all of this consumes. We're starting to get out of cheap territory.
I suppose what I'm trying to say is: If you have a lot of data, hard drives are only cheap if you don't care availability, integrity, and longevity.