This is already factoring in 2x storage, so you just put each half in a different location.
> and they likely have some kind of offline archive too.
Okay so 2 million.
> Then there is the running cost to power all those drives.
Yes, that's one of many things that adds on top of the raw drive cost.
> Also, when you have that many drives, you're going to be constantly replacing dead ones.
It's not actually that bad! If you cram ten thousand drives into a few racks, your drive failures per week will be in the single digits. You could have someone come in once a month.
> I have a feeling this is an expensive operation!
It is, but my core point is that it would be expensive even if the drives were free. You need a certain amount of server horsepower and the physical resources to support it. This implies that you're not getting to stick 60 drives in a single chassis, so you need more labor as well.
There was an estimate last year of $1500 for IA to store and serve a terabyte of data indefinitely. I'd love to see a breakdown of that number, how much goes to day 1 hardware, how much goes to year 1 hosting, how long that hardware is expected to last. But even the day 1 cost is definitely a lot more than $50 for 2TB worth of hard drive.
My bad, I missed that!
> It is, but my core point is that it would be expensive even if the drives were free.
On reflection, and after reading your reply, I agree :)
> There was an estimate last year of $1500 for IA to store and serve a terabyte of data indefinitely.
I assume that's $1500 per year?
Total, not per year. Even S3 would only cost $250 a year to store a TB redundantly, and you could host 20 of them off a $200-per-month server or something.
Backblaze gets you a bit more than one copy for only $60 a year.
Then there is the running cost to power all those drives.
Also, when you have that many drives, you're going to be constantly replacing dead ones.
I have a feeling this is an expensive operation!