Hacker News Reader

Comment by giancarlostoro | Hacker News Reader

So is this primarily for static websites? I don't see dynamic websites going too well with this system.

Edit:

If someone like GitHub supported this for GitHub pages it would be a great step forward for this as well.

With IPNS, which allows the creation of links whose destination can be mutated by whoever has the private key, you can implement quite a lot of interactivity. You won't ever be able to implement facebook over this platform, but you can replicate all of its functionality by using a tumblog-like architecture. Facebook's design is inherently centralized, and therefore doesn't really fit into the world of IPFS.

The idea is that every user owns their own content, not some central organization. Every user maintains a log of all of their actions, like an RSS feed. Every time they make a status update, that's an item. When they comment on someone else's status update, that's an item. Even when they just like something, it's an item. Each update of the action log would also have send ping to all friends, over some kind of distributed IM like tox. They already need an IM connection like that for messenger.

Given that information alone, anyone can reconstruct the activity of all their friends into a news feed, and post new content. You'd need a lot of client-side leverage, and the ability to upload to the IPFS network from the client, but all the dynamic information is there.

And there you have a fully distributed facebook clone. More resilient and privacy-sensitive and monopoly-averse than even tools like diaspora and statusnet.

How does friend recommendation work in a decentralized model?

Isn't friend recommendation mostly stuff like "people who follow you/you follow also follow X"? Maybe going one or two more follow hops, but not so far out that you need a global view of the social graph. If users of a decentralized network publish their following and follower lists, it should be pretty easy for a local crawler to check those hops.

You have friends. A script can check your friends, and make suggestions based on that.

You won't have the full graph that Facebook does, but does that really matter? I suspect the best predictors of who your friends are would be people already in your network.

> Facebook's design is inherently centralized

Not really. There's no reason you couldn't implement Facebook's core functionality with personal servers containing a profile at each node and a pub/sub (or even just pull) system connecting each node with its friends.

So, diaspora*?

https://diasporafoundation.org/

Unfortunately it never really caught on.

Yes, I believe diaspora was a social network of this sort. But it's just one example, and not an indictment of decentralized social networking in general. Your address book + SMTP is a decentralized social network that caught on just fine. The only difference is that it was designed and built before corporate interests were the driving forces behind new standards (or lack thereof).

Actually, this is exactly what I'm looking for. I want to be able to control my photos and other content, and share that with family and friends.

All data is encrypted and my server can be hosted anywhere (just as simple as setting up a WordPress instance).

My own "news feed" are a combination of all of the other nodes I'm subscribed to.

My problem with Disapora is that they push the "sign up to someone else's pod" rather than "install your personal pod".

It just feels like a multi-centralised model. Rather than one Facebook with my data, now there are 311 "Facebookesque" Diaspora Pods (as of today).

As much as people hate the idea, a WordPress like (PHP) application as a "personal server" would be successful. There are lots of hosting companies out there where you can get up and running for peanuts.

My aim would be for a simple install:

- Download software and upload to hosting company

- Enter MySQL auth details

- Choose local filesystem or third party (S3, DropBox, Google Drive, MS SkyDrive) with OAuth as the backing store (files stored encrypted).

- Finish

Then I would be able to subscribe to other friends and family personal servers, just like RSS feeds. Friends can access my own personal server through OAuth and leave comments on content. Those comments are echoed back in my private feed which is displayed in their merged news feed on their own app instance.

The technology is already there. It just needs someone who is damn good with PHP!

We're heading towards this with Peergos (https://peergos.org source at https://github.com/ianopolous/Peergos). Encrypted, self-hostable file storage, sharing and social network. We're considering using IPFS under the hood for fragment storage/finding. You'll be able to either use your own disk, or enter credentials for a cloud provider. It's still very early days yet though.

The reason why they get people to sign up at other people's pods is that it's easier to get users that way. They are struggling to get users regardless and that is their biggest flaw.

That sounds a lot like the federated media hosting that GNU MediaGoblin is creating.[0]

[0] http://www.mediagoblin.org/

I'm thinking about the discovery problem and how this relates to distributed technology now. File storage looks solved but creating content relationships like lists of similar files sounds much harder and still reliant on a authority that curates and filters the spam. Lost data in IPFS will be "buried treasure" in that nobody has the link to it anymore, but it still exists in a literal sense.

Interesting to think about.

There are plans for a (distributed) search engine: https://github.com/ipfs/archives/issues/8

A few questions about IPNS:

- Is it possible to, say, log into my bank account, see my current balance, and set up bill pay?

- Can I order a pizza over IPNS?

- Can I do realtime chat?

Also, this article says we could use Namecoin and not rely on ICANN for DNS, however, you still need ICANN to distribute ip address to get connected to the internet in the first place. Or am I missing something?

1) 2) 3), yes, it is still possible to directly connect to specific peers through IPFS, see http://gateway.ipfs.io/ipfs/QmTkzDwWqPbnAh5YiV5VwcTLnGdwSNsN...

I think that it should be possible to build hybrid centralized/decentralized sites with IPFS, with static public content being fully decentralized and sensitive information going through direct tunnels. Of course, the more decentralized your system the better.

As far as the IP system, yes we are unfortunately still reliant on ICANN being sensible about IP allocations, but the DNS system can be fully deprecated. I do know that there is a project out there to redo the IP system in a decentralized fashion but I can't recall its name right now.

> project out there to redo the IP system in a decentralized fashion

https://github.com/cjdelisle/cjdns

That's the one. Juan mentioned it in one of his talks, said that it would be possible for IPFS to sit on top of that and use it as the network layer.

The first two cases you mention are about communicating with a specific real world organization. You can implement it using ipns, but it's inherently centralized.

The last one is a no. Well actually yes if you just poll a file, but that's not practical on a large scale. You need an external communication channel to implement that. It's possible to have that be fully distributed as well though, as in tox. You need this external communication channel to implement a lot of things anyway, so it would probably be part of the platform that forms around the idea of a distributed web.

IPFS is designed with mutability in mind-- think of it like git branches which advance to newer versions. (though you can indeed get rid of version history in that pointer if you want). So, the short answer is "yes, IPFS will work with mutable content just fine", though _how_ the mutability works is a harder, deeper question.

I think the best sources will be the paper, the talks on the website, and the FAQ:

- https://ipfs.io/ipfs/QmR7GSQM93Cx5eAg6a6yRzNde1FQv7uL6X1o4k7... - https://ipfs.io/ - https://github.com/ipfs/faq/issues

They could take an approach similar to how Freenet does dynamic things. See here for details: http://draketo.de/light/english/freenet/communication-primit...

Freenet has a similar immutable datastore but chat, microblogging, etc have been built on it.

> So is this primarily for static websites?

More work would be needed here perhaps. But decentralized apps should happen one day. As someone without real technical knowledge in security and decentralized platforms, I'd love to hear from some on HN where we really stand on decentralized apps and security.

- Public data. Distributing, searching and querying is not a problem.

- Restricted data. You'd be distributing encrypted data to peers. Now, if we have a way to send an encrypted query to the peer and have it run on encrypted data without the peer being able to understand the query or the data, we'd be set perhaps.

- Obviously this would need application architectures to change. But that's solvable.

> Now, if we have a way to send an encrypted query to the peer and have it run on encrypted data without the peer being able to understand the query or the data, we'd be set perhaps.

Is that possible? It doesn't seem to be "N versus NP" but it seems similar. How can a computer find a correct answer to a question it doesn't know?

I suppose you could send off small units of processing across many nodes, such that none of them knows the full result set...

Since the private data and the query are from the same user, they could share some mathematical connection perhaps, which would allow a query to be done and the results sent back (results which only the user can decrypt).

Google shows up plenty of articles on the subject, but I am not an expert to know where we stand on this.

Interesting. So you'd perform operations on encrypted data, and send back an encrypted result that's useful yet different from the original encrypted data, all without ever performing decryption?

It seems possible but extremely tricky.

https://en.wikipedia.org/wiki/Homomorphic_encryption

My understanding is that it is possible, and Turing-complete systems can be built, but they are slow and unwieldy, requiring a large amount of cyphertext to encode data (on the order of 4KB per float, I seem to recall).

It's still up for debate whether or not such techniques will eventually be up for general distributed computing, e.g. distributing a VM instance over multiple distributed servers; or just for specific use-cases, e.g. bitcoin-esque distribution of small transactions, etc.

This is called "homomorphic encryption". There's a lot of research in this area, but we're far from being able to efficiently perform all computations.

I don't really know about restricted data, but AlexandriaP2P is doing pretty interesting work with storing IPFS magnet links in a blockchain.

How dynamic? Resources can be updated to point to new immutable objects. For instance, as an operator of a website, I can sign the resource `/comments/10189265.json` as referencing an object with a particular hash, and then when there are additional comments, change the object it references. How fast things like this can update are dependent on the implementation, which I haven't read.

This is also the qualm I'm having with it... it's great for HTML files, but dynamic websites go right the window.

I believe they have an "IPNS" protocol which is intended to handle mutations.

Any links, please?

You can read section 3.7 of their paper for an overview: http://static.benet.ai/t/ipfs.pdf

I imagine you could go through their codebase to try to see how far this has come along: https://github.com/ipfs/go-ipfs

According to this, it looks like the implementation is currently in progress: https://github.com/ipfs/go-ipfs/blob/master/dev.md

Hope something in there helps you.

Reply on news.ycombinator.com