In most schemes I'm aware of, the nodes are storing useful data. But perhaps you are suggesting that they store arbitrary data, e.g. deterministically generated data?
My colleague and I wrote a paper about something like this a while ago. We called it "proof of capacity." The basic idea is that you deterministically generate a bunch of hashes based on some seed. Then, at a later time, you are "challenged" to return one of the hashes. Pretty standard so far, but here's the trick: you aren't asked to return, say, the hash at index i. Instead, you are challenged to provide the hash lexicographically closest to some random hash X. This obviously requires sorting the generated hashes, which is why you can't just regenerate them on the fly. And you can get a rough estimate of the node's capacity by seeing how close their hash is to hash X: the more capacity, the more hashes, and thus the closer their hash will approach X.
The core premise is that you fill your drive with a special type of random data (can't use useful data, therefore it's still wasteful!) that you then perform checks on regularly.
There's Burstcoin, which best I could tell has some substantial flaws, though I'm not sure anyone has ever taken the time to prove it. (Burstcoin is small and therefore not super worthwhile to break).
Then there's Spacemint, which is not yet complete but has some strong progress. I've gotten to talk to them, and best I could tell they were aware of most of the major weaknesses of their protocol, which generally is a good sign. At that point they still required an external source of secure public random numbers, which is not an easy problem to solve without introducing trust or PoW. But maybe the protocol is further along by now.
https://www.google.com/url?q=https://eprint.iacr.org/2015/52...
You are missing the elephant here. POW doesn't consume massive amounts of energy. It consumes the amount of energy the network will pay for. If RAM becomes POW, you'll see hundreds of factories stacked with RAM bars.
Seems it would use a lot of bandwidth or computation when calculating hashes.
I was thinking something like write a bunch of random numbers to disk, and prove that you have the ability to retrieve any of them at will.
This is obviously not a complete solution. I was asking if anyone knew of any such system.
A positive, not negative, feature of the system is you must be able to rewrite all the storage each time. So the faster your can write the more blocks you win, but also the more storage you have the more blocks you win (and it's a tradeoff between them).
It's an interesting idea.
The result so far has been really potent. On the Sia network today, raw storage (pre-redundancy) costs about $0.30 per TB. Performed intelligently, you only need about 2x redundancy, putting the total cost at less than 1/10th of Amazon glacier.
I project that as the network grows, demand will catch up to supply more and the price will settle closer to $2 / TB / Mo for raw storage.
I can answer any questions.
This would force people to have physical objects (RAM or HDD) to mine, but doesn't consume the massive amounts of energy that POW mining consumes.
(i.e. you only need the storage to mine a block, not longterm)
The way I'm envisioning it each "tick" randomly X blocks are picked, and the more storage you have the more likely it is you will be picked (potentially more than once).