Comment by pradn | Hacker News Reader

Comment by pradn | Hacker News Reader

Parent Full thread

Current shadow libraries (zlib, libgen, scihub) suffer from centralized data hosting and opaque librarians/custodians (who modify metadata and gate inclusion/exclusion of content). We already have the tools to solve this.

1. Files are stored in a distributed fashion and referred to via their content hash. We already have IPFS for this.

2. Library metadata can be packaged up into a SQLite DB file. The DB would contain IPFS hashes, book names, authors, etc.

3. Teams of volunteers assemble and publish the library metadata DB files. There can be multiple teams, each with their own policies. The latest library files can be published via RSS. Each team can have their own upload portal.

4. A desktop app can pull multiple RSS feeds for multiple libraries. The libraries can be combined together and be searched easily on the client side. Users can search for content via the latest library metadata files, locally on their desktop. Content can be downloaded via IPFS.

5. The desktop app can also double as an IPFS host, allowing users to choose specific files to pin or simply allocate an amount of space for the purpose (100 GB, etc). There could also be servers that aggregate pinning info to make sure no gaps are there.

5. For ease of access, people can run websites that preclude the need to setup your own desktop app / download libraries.

6. Library teams can publish metadata DBs and content via torrents, too, for long-term/disaster-recovery/archival purposes.

This would be a true hydra. No one centralized team, no reliance on DNS. If one team's library set up goes down, you can use another's.

1. yes, ipfs could solve that, but it relies on people hoting content. Previous examples of content-based addressing showed that little-accessed content tends to disappear as nodes go offline over the years. This would need to be solved, and I think the only way to solve it is to have a battery of centralized ipfs servers mirroring each other, which defeats the "fully distributed" setup

2. this would also need to be hosted and could be taken down. You'd need to mirror this too, but that's a simpler problem to solve (gigabytes instead of terabytes)

3. the upload portals and the RSS feeds would, again, be centralized or would have to change so regularly that they become impractical

in the end you would end up with a dozen (a hundred? more?) different z-libraries, which would make it actually worse from a preservation standpoint, since only the most popular content would be shared, libraries that focused on rare/exotic/fringe material would be endangered of being lost since they have fewer volunteers/mirrors/seeds/...

Also, freenet and other projects already showed that end-users allocating some storage and using that to spread data around is not an easy problem, the fluctuation in end-nodes is so big that it slows down the entire network to a crawl. I'm not sure this problem has been solved yet.

1. You're absolutely right that IPFS alone isn't a good way to guarantee durability. I think you'd need a second level of archiving done in bulk. Thousands of books collated into various torrents that people can help seed. This has already been done for LibGen. IPFS does provide a common hash space for everyone to rally around, and does make it easy to download single books. It's also easy to speed up downloads by using an IPFS gateway - the actual protocol is slow. I don't expect most users to actually pin files or anything.

2. Well, users and librarians need some way to find each other. That's true in any system. And that communication medium can allow certain kinds of attacks. (website on the public internet, word-of-mouth, Telegram groups) If all someone needs is an IPFS hash of a recent library metadata DB (a SQLite file), any way of communication will suffice. I think this approach allows for centralization (sure, keep a website up as long a the authorities don't care) but also gracefully allows for all manner of de-centralization (use any of the above methods to distribute the metadata DB).

3. Any many-to-one system with curation (librarians) will have weak points. The idea is you can set up upload portals across any communication medium (a regular website, a dark-net site, a Telegram group, email) - and the libraries take care of collating the information. The social grouping is what matters more (libraries vs uploaders vs downloaders) - and we want to make it tech agnostic and, therefore, more resilient.

This system will be stable I think, for two reasons:

1. Network and branding effects will naturally create a few big libraries. People will use the familiar, useful libraries. See how few torrent search sites took up the bulk of traffic, back in the heyday of torrents. Most users will probably use a website, and the ones that are easiest to use will probably get the most traffic.

2. The resilience of the system is necessary only once in a while. A set of libraries will emerge, there'll be enforcement actions and they might break apart, and then new ones will pick up their pieces (easily bc the metadata and content is open). So we want to provide the open-ness for this to actually happen.

The core problem with IPFS and friends is that the vast majority of Web access these days occurs on battery-powered devices that need to sleep their connections often and for long periods or battery life will plummet. End users aren't going to accept even 10% worse battery life (and it'd likely be worse than that) just so they can participate in "the swarm" and have higher latency and slower transfer on all their requests.

You could keep the metadata DB in IPFS too, along with the books. The only thing that needs to be regularly republished is the CID of the metadata DB, which fits in a tweet.

In fact, it fits in a tweet with enough room for steganographic deniability hijinks. You could publish a browser plugin which transforms tweets into IPFS CIDs according to a non-cryptographic hash algorithm. That way the cleartext of your tweet is not takedown-worthy, nor is the plugin, but the two together let users transform the seemingly innocuous tweet into the metadata DB update they need.

That's a great idea, and also amusing. :)

It's amazing that we can refer to data in a globally-unique way with small content-based hashes. Hash collisions aren't usually worth worrying about.

Another benefit is that its easy to store large numbers of hashes with basic metadata.

SHA-256 hashes are 32 bytes. If it takes 512 bytes on average to store author/title/publish-date/ISBN, then the hash is a small part of the total per item (though not well-compressible.) You can store the info for 2 million books in a megabyte.

Shadow librarians can also publish curated collections of books. I know a guy who tried to do this in a systematic way for college-level history textbooks covering a wide swathe of the world's history. The entire catalog with metadata and hashes is probably only a few hundred thousand KB.

The tricky part is finding a balance between the blockchain approach, where everybody pins every item (wasteful) and the other end of the spectrum where people only pin the things they're thinking about right now (lossy).

There's some middle ground where we coordinate who pins what in a way where there's just enough enough redundancy to not worry about it disappearing, but not so much that any one of us is bearing an unnecessarily large burden.

I don't think we've quite figured that part out yet.

You bring up a great insight - that blockchain/pinning are on opposite ends of the spectrum of what one can choose to store, if one participates in the system at all.

We do have a good example of this already I think: torrent client peer lists. When participating in a torrent, we can see the current availability of the data in the swarm, displayed usually as a bar visualization, with each pixel being a chunk. The darker the chunk, the more peers have a copy of that chunk. The result is also summarized in a number capturing the general health of the swarm wrt hosting the torrent.

All we need to do, then, is to have a mechanism where 1) the client has a list of items to pin 2) the client uses this existing swarm-tracking mechanism to figure out which files need more hosts 3) the client picks the best ones to host given the available space/network constraints. One can be smarter than just picking the lowest-seeded file. If a host is known to have files consistently, but is offline for a few hours a day, the client can be smart and not worry about immediately getting those files, perhaps spending available resources on less-seeded files.

This is possible with current technology. We can do a simple version of this via a torrent client plugin, reading the list of files from RSS or a special text file.

I've seen communities do this manually actually. For hosting SciHub torrents, the community made a web page that showed the current number of known seeds per torrent. Users were responsible for picking whichever ones, usually the lowest-seeded ones, and seeding them. We can remove this tedious and error-prone work.

Doing this per IPFS file will probably take up too many resources. Perhaps we need a standard IPFS<->torrent correspondence. Something as simple as a text file in the root of the torrent file structure, a file that maps IPFS hash <-> file inside the torrent. This way an IPFS swarm and a torrent swarm can work together. You get the easy retrieval of IPFS and the increased durability of torrent swarms.

> Doing this per IPFS file will probably take up too many resources

I think we can come up with some scheme where it doesn't have to be centrally hosted. Like if my public key is 18 mod 256, then it's up to me to pin all of the files I rely on whose CID is also 18 mod 256.

If you've got thousands of users doing this, each one of them has to bear only 1/256th the burden.

I imagine incentive schemes where we keep track of which peers have gotten which files from our nodes, and then later randomly check to see if they're doing their part and pinning what they got from us

We'd all put $5 into a pot at the beginning of the month, and at the end of the month we'd share our data re: who seeded and who leeched. Maybe the bottom 10% leechers get no money back, the top 10% seeders get $10 back and everybody else gets their $5 back.

So it's like it costs $5 to access the first time, but if you leave a node running (like maybe you have a phone with a cracked screen that you just leave plugged in) then you'll never have to pay that $5 again, and if you're lucky you'll get $5 from a leecher.

Of course it doesn't make sense for all files anywhere, but in context with a project, like Z-library, where somebody is curating the files that are in scope. Otherwise an attacker could flood the network with noise that hashed in such a way that it affected only their target (and their target's 1/256 slice of the community).

… taking this a step further… you could place the message you speak of tweeting into an ‘oh by’ code[1] … and then just chalk it onto the sidewalk.

Now passers-by can receive the message, time shifted, without the Internet.

[1] 0x.co

I'd like a stencil that can be configured to display an arbitrary QR code so you can do it with spray chalk.

Actually, you can do another cool thing, too. The IPFS ecosystem has IPNS, which generates a public/private-key-back hash-name, which can point to another IPFS hash. So this way, a shadow librarian group can have one IPNS hash that always points to the latest catalog.

Right, I had forgotten about IPNS. I remember some thread (a while back) where people were complaining that it was terribly slow, so I wrote it off.

Now that I think about it again, I'm not sure what you'd use IPFS for that would require fast IPNS resolution anyway. Having version of a library that's an hour old is... just fine actually.

https://annas-archive.org does a lot of this, and has been my go-to for books for a while now.

Yes! Anna has been active in helping duplicate Zlib, among other efforts. No one person can do it all. We need to pitch in!

It has already done, you will be glad to learn:

https://bafybeigpp6mtsmjngaqscfkjwzivbptt4ui5yb7uih6qe43obof...

For me it has worked in Brave, Chrome, Firefox and Safari Tech Preview, with or without installed IPFS and IPFS Companion (with IPFS it works much better). Haven't worker in Safari in IOS and non-Chrome browsers on Android.

Not very fast and sometimes it requires page reloading but overall impression is awesome.

Thank you! This looks promising!

Guys, you already have this one named Nexus. Even for torrents.

https://bafybeigpp6mtsmjngaqscfkjwzivbptt4ui5yb7uih6qe43obof...

You must learn about them.

An alternative that works today is #bookz on undernet.

Reply on news.ycombinator.com