Multiple nodes per endpoint

[shishkin]

I'm trying to implement a use case with iroh where a single app instance can have multiple peer identities (like user profiles). Looking the Endpoint and NodeAddr source code it seems that there is a 1:1 relationship. Would I have to build peer identity in my own protocol on top of iroh? Is there anything else in iroh that I've missed that could allow adding more NodeAddrs to a running Endpoint?

[flub]

Think of a NodeAddr as a SocketAddr: if you want to be addressable for different identities on the same socket then you need to do identity on your own protocol. If however you want to have different networking endpoints then you need to bind multiple endpoints.

[shishkin]

Thanks @flub. How severe is the overhead of an additional endpoint? How much routing data (or something else?) is it going to accumulate/duplicate?

I was thinking to have multiple endpoints for the sake of privacy, so new identity is not correlated with the same NodeId. But I guess the NodeAddr will give the IP address away for correlation anyway? Does Iroh have any plans for privacy-preserving routing like Veilid?

[flub]

There is a certain amount of overhead with using multiple endpoints because they will be doing similar work. So yes, that does have a downside.

Regarding privacy: iroh's primary aim is to establish p2p connections. This is very different from being a mixnet. The peers you interact with are going to know your IP addresses. Improving privacy has been something we not yet prioritised, reliable connectivity being more important. But likely we'll move to improve this at some point. For example we can make sure to only expose IP addresses to peers you want to talk too, this is mostly (or even entirely in some circumstances) the case already, but I can't promise this right now since it's more a side-effect and not been specifically engineered. Also some users have asked about relay-only and double-relay versions which would also allow you to improve IP privacy at the cost of more relay traffic. But none of that is going to happen soon.

Regarding NodeId's, what many users do is not safe the NodeId to disk, then you generate a new one every time you start up. Depending on your usecase that might be fine. This is even more likely if you rely on identity/authentication in your own protocol.

[shishkin]

Yeah, generating fresh NodeId for each session makes sense.

Could you please help me quantify (to orders of magnitude) memory overhead of an endpoint?

[flub]

I'm not really sure, you'd have to test this out yourself for your scenario.