lucash-dev
commented
1 year ago @mikedilger first of all, thank you very much for your thoughtful response to my article.
Let me try to answer some of your points:
There is no assurance that your message will get through potentially naughty relays. But in thinking about it, how could any such assurance ever arise in any other protocol? Someone could pull your Internet connection. How would a protocol get around that one? It seems intractable to me, and too high a standard to hold anything to.
The point is that if someone pulls your Internet connection, that fact will be pretty obvious, and you'll have a chance of trying to find a different connection (ultimately someone could nuke ISPs and then it's all over, but that extreme isn't usually part of any definition of censorship-resistance).
In Nostr it isn't clear how an author could go about detecting their events are being censored to a subset of users, so they can't try and find a better relay. That means the relay (or set of relays your self-hosted relay is connecting to) is a TTP (trusted third party) in the same way Twitter (or a set of platforms such as Truth, Gab, etc) are TTPs. Lack of Sybil protection against relays makes it even worse. It's essentially impossible to know if you're even connecting to different relays or just one relay with aliases.
Compared to Bitcoin, if you're being eclipsed by nodes you're connected to you'll see a significant drop in hash-rate, so you can try and find better peers. Comparing to Tor, while privacy requires at least one node to be honest, censorship can be instantly detected (you can't reach the other end).
I'm sure it's possible to create a protocol for broadcasting messages (which is essentially what having a blog is) that have similar guarantees as Bitcoin and/or Tor. One obvious solution is throwing everything into a proof-of-work blockchain, but that has many practical issues. There are some theoretical solutions provably censorship-resistant, but they also have practical limitations. I believe other solutions with different trade-offs that are more practical can be found.
Consider the case of a user owning a relay, and posting their own posts to their own relay, much like a website. And consider other people following that person by pulling from that relay, much like RSS. Under this model, the trust issues between relays and users which you have highlighted vanish.
That would mean Nostr isn't much improvement over just running any server supporting RSS, and that topology can be pretty much perfectly censorship-resistant if everyone uses Tor. Even without considering limitations in Tor, that isn't scalable though -- which is one of the main points things like Twitter were created in the first place.
If everyone who follows you need to pull events from your server, then the only way you can scale to something like millions of followers is by having a server that can handle that much traffic -- and paying for lots of bandwidth. Even if you add micropayments for fetching each event, it's not obvious how individual users could deploy and manage a server that can reach that scale -- they'll definitely have an incentive to just host everything on AWS and similar platforms (and even that isn't easy for individual users). Then again, you're back to the same trust model as having accounts on a few platforms like Twitter and signing each with your PGP key.
The fact that messages cannot be altered, that users can move to different relays without losing their identity, and they can spin up new identities, gives a modicum of censorship resistance, but not in the formal way you have analysed it.
I don't think it adds much that wasn't already possible with existing solutions like Twitter+PGP, RSS+PGP or ActivityPub+PGP. It might be more elegant than these alternatives, but it isn't more censorship-resistant.
mikedilger
The security of a protocol or of a deployment depends quite a lot on the security model at play.
The current deployment of nostr, and the current behavior of most of the clients, brings about the vast majority of the issues you have highlighted. But few of them are directly attributable to the protocol, which is topology agnostic.
Consider the case of a user owning a relay, and posting their own posts to their own relay, much like a website. And consider other people following that person by pulling from that relay, much like RSS. Under this model, the trust issues between relays and users which you have highlighted vanish.
The current set of clients and deployment is not scalable and has many issues. I keep harping on about it, and fiatjaf is fully aware of these issues and says he also has been talking about it incessantly. I want clients to follow users at whatever relays those users happen to post to, and not expect their posts to be at some small set of client-configured relays. What they are doing now is akin to hoping to find Richard Dawkin's blog on Steve Bellovin's website. That's one of the reasons I wrote the gossip client, to show how it could be done differently.
That being said, nostr is supposed to be such that users don't have to trust relays. But clearly all you have said about censorship resistance is true. The fact that messages cannot be altered, that users can move to different relays without losing their identity, and they can spin up new identities, gives a modicum of censorship resistance, but not in the formal way you have analysed it.
There is no assurance that your message will get through potentially naughty relays. But in thinking about it, how could any such assurance ever arise in any other protocol? Someone could pull your Internet connection. How would a protocol get around that one? It seems intractable to me, and too high a standard to hold anything to.