FWIW it seems like NixOS at least tried or is trying: https://blog.ipfs.tech/2020-09-08-nix-ipfs-milestone-1/
Every time someone downloads a book from a shadow library like Library Genesis, it's through IPFS most of the time, often via an IPFS gateway such as Cloudflare's IPFS gateway, so you don't even notice it's using IPFS. These shadow libraries have millions of users per day, especially academics.
You bring up a valid point though. From my experience, IPFS suffers from a tragedy of the commons issue. It's performance is generally terrible because the free providers are overloaded.
> Every time someone downloads a book from a shadow library like Library Genesis, it's through IPFS most of the time
I would like to see evidence this is actually true. My theory is that IPFS's "success" in hosting Libgen is roughly turning ICO proceeds into free hosting. Paying for dedicated servers might have been more cost efficient and, given the performance issues, more robust!
I really had high hopes for it, but I realized that all I really want is an object storage with content addressable urls.
In fact, while F2FS is a ludicrous example, this happens regularly with filesystems. The core design of filesystems is still built assuming that a block device will be behind it. IPFS is the same way, just s/block device/blockchain/.
- You can have files on IPFS and no blockchain whatsoever "behind it".
- IPFS itself makes no use of blockchains as an abstration for data storage or file organization
- If all the blockchains networks and nodes disappeared overnight, no IPFS nodes would be affected.
I don't know if you actually used any of that or just going by hearsay.
There's a separate "layer on top of IPFS" called Filecoin that does involve a blockchain. (From a technical point of view, it's not really a layer on top, more of a re-implementation of IPFS with a blockchain.)
I've worked on IPFS and know many of the key people at PL. They really are not blockchain people or cryptobros. They are much different from anyone else in crypto I have encountered. But ultimately they have to fund their activities somehow, and this is the reality of where the business potential is in P2P circa mid-2020s.
This was one of the things that made IPFS a non-starter for us. We ended up grafting Hashicorp Vault into kubo (the go-ipfs implementation) so that we could use IPFS and have things like detetes and access revocation that actually work.
https://guide.fission.codes/developers/webnative/file-system...
We have block level access control for example: https://peergos.org/posts/bats
As well as E2EE, sharing, trustless servers and more [1].
There is a Java implementation of the modified bitswap for this in Nabu [0] and a Go one in ipfs-nucleus [1]. With both of these you can use whatever auth verification protocol you like.
When it comes to FOSS, I personally don't understand why something like a package manager isn't P2P by default. It feels very aligned with the hacker culture - a la "A Declaration of the Independence of Cyberspace". Virtually nobody uses it, so the solutions are half baked, clunky and not integrated with every day workflows (ex: browsers don't support anything P2P). Something like libcurl can't pull a torrent from the web
My (possibly wrong) feeling is that IPFS is meant for smaller files, and torrents for bigger collections. So with the example of a package manager, you could download each individual package via IPFS, whereas a torrent would make sense if you wanted to download an archive of them all.
Not to say torrents couldn't work, but it just kinda feels like not the intended use case.
(The above feeling come from reading about IPFS a few times throughout the years and toying with it for a while).
Packages are a lot of small files, and not everyone is interested in them all. Torrents are good for large content that everyone wants. The two are in opposition and there needs to be a balance: do you need to make One Unique Torrent with everything, but then peers have to individually ask each and every other peer if they have the content they're looking for, making it unscalable ? Or do you make one torrent per package, meaning peers have to track thousands of torrents ? (That's the direction IPFS took, which is why it's extremely inefficient and resource-heavy). DebTorrent has made the choice to go in the middle.
The real problem, to me, is that content-addressed content is the anti-privacy protocol: since you ask everyone "do you have this content", you're basically telling everyone what you want. In the case of installing packages, this means telling everyone what the version of your packages is, and is potentially a security issue.
But the real "problem" is that it fails at being much more better than the existing solution. Bandwidth is cheap, storage is cheap, so hosting everything on a few independent but centralized servers is good enough.
Netflix uses IPFS. [0]
[0] https://blog.ipfs.tech/2020-02-14-improved-bitswap-for-conta...
I think you might misunderstand how IPFS works, or possibly confuse it with Freenet. When you use IPFS, nothing is automatically distributed, unless they happen to know the hash of the content you've added locally. It's not until someone starts to explicitly request data from you, that you start sharing anything.
Assume A is the copyrighted work. Now construct a file of random bits and call it B. Then assuming A=B^C, then you can solve C=A^B (where ^ is the XOR operator). Both B and C appear to be completely random.
Assume you are sued because C is found on your harddrive and if you XOR it with B (found on someone else's harddrive) you get copyrighted work A.
To construct your defense, take a work in the public domain, and call it P. Compute D=C^P. Now tell the judge that there is a file D on a friend's harddrive, that allows you to reconstruct P.
Alternatively, tell the judge that you are researching random numbers and C is just a file you work with. It is not illegal to have random numbers (bits) on your harddrive.
I mean, the XOR thing is a bad idea, because obviously at least one of those files is violating copyright. Trying to feign innocence is not engineer logic, it's failed logic.
But Aereo I'm still mad about. They saw that distribution is what cable companies have to pay lots of money for, so they removed distribution. It should have been fine.
The effect of the decision is that it's legal for a person to set up a remote antenna for personal use, but it's a copyright infringement to pay a company to set it up for you. Or, perhaps, that it's possible to pay a company to do so but only if it's your idea as a one-off. Either way, that doesn't make any sense as a copyright issue.
And after the supreme court decision, they tried to pivot to legally being a cable service... only to get denied recognition as one!
My point in the GP comment was more that if the entire justification for your technical architecture is to make an end run around some law, you have to consider the possibility that a judge is going to see what you're trying to do and may have a different interpretation. There's a big human element in the law that I think engineers, especially "code as law" blockchain people, often overlook or misunderstand.
Maybe in the future we'll have AI judges and AI lawyers and your case will be decided in 100 milliseconds. However, that's not what we have today.
The point with the XORing is that you can't blame one person for the infringement.
Imagine twin brothers. One of them steals a painting and is caught on camera. The judge knows that one of the twins is guilty. However, he can't convict either of them exactly because of the human element in the law.
I love the idea about data being addressable by hash, but I don't feel IPFS has actually delivered anything meaningful in that area yet.
And with ipfs-search.com shutdown, there is not even any way left to explore what is actually on the network now.
And technical issues aside, there is also the legal problem that IPFS conflicts with copyright. Redistributing anything is illegal by default, unless somebody gives you permission, IPFS provides no means to track that permission. You can't attach a GPL or an author to a hash.