[tor-dev] Scalability or Onionbalance for v3 ephemeral/ADD_ONION services
George Kadianakis
desnacked at riseup.net
Mon Jul 26 10:20:35 UTC 2021
Holmes Wilson <h at zbay.llc> writes:
> Hi George,
>
> Sorry for the slow reply here! Just getting back to this.
>
>>> For our application (a messaging app) it would be super useful to get the
>>> full list of known online (or recently seen online) onion addresses in
>>> possession of some frontend key. This would let us use onionbalance for
>>> peer discovery instead of blindly trying the set of all known peers, which
>>> won't work well for large groups / large numbers of peers.
>>>
>>
>> Hmm, can you please give us some more details on what you are looking
>> for? What is peer discovery in the above context, and what do you mean
>> with "full list of ... onion addresses in possession of some frontend
>> key"? I'm asking because the frontend key of onionbalance is also the
>> onion address that users should access.
>
> Our context is we are building a Discord-like team messaging app where peers are connected to each other over Tor, via onion addresses, rather than to a central server. So each user connects to a few peers, and messages travel across peers on a gossip network, and there’s a mechanism for syncing messages you missed, say, if you went offline for a bit.
>
> One problem we have is, when a new peer comes online, how do they know which other peers are online? Right now, they can try all of the peers they know about, or perhaps try recently-seen peers. But if there are hundreds of peers and only a few are currently online, it will be necessary to try many unreachable peers before finding one who’s online. So that’s not ideal.
>
> One solution to this would be for each online peer to host the same onion service, using a shared key, in addition to their normal peer onion address. And at this address they could return a list of peers they knew were online. So a user would just have to connect to one address, at which point the Tor network would connect them to some online peer, and then that peer could tell them about other online peers. The problem with this approach, as pointed out by folks on this list, was that all those peers would have to really trust each other, since any one of them could go rogue and host malicious information instead of the peer list, gumming up the works. I’m not sure this is a fatal problem, since it would still *help* in cases where there wasn’t a malicious peer, and users could still fall back to the slower method of trying every peer.
>
> But what I’m wondering is whether there is any mechanism for a bunch of onion addresses that *don’t* completely trust each other to share a “meta” onion address on the Tor network, such that when the user looks up that identifier instead of getting connected directly to whatever content one of those onion addresses is serving, they get a list of all onion addresses that hold the keys to the “meta” address.
>
> It’d be like asking Tor, "show me a list of all onion addresses that have registered this meta address.” Sort of like asking, “show me a list of mirrors for this address…” at which point the user could try connecting to one or more of them, but would not have as serious problem if one of the sites went rogue and started serving useless content.
>
> This is a bit of a long explanation, and my guess is that there isn’t anything like this and that the above scenario isn’t common enough to be worth targeting, but I was curious if anything like this had ever been discussed.
>
Hello Holmes,
I don't know much about these kind of P2P protocols, but my intuition is
that this "get list of online peers" should be handled on the gossip
protocol layer, and not on the Tor layer. As a naive strawman example,
each peer can keep its own list of online peers and return it when asked.
I feel like the idea of "connect to meta address to get list of online
peers" is kinda the same as "ask any peer you can find for the list of
online peers". That's because with the meta address idea you don't have
a way to know whether the meta address result is a trusted peer; in the
same way that you don't know whether the peers you get through gossip
are trusted peers. This means that in either case you will have to
handle malicious nodes somehow.
In any case, the "meta" address idea is not handled natively by Tor
right now. You could in theory do it by having multiple peers share the
same private key, but I don't know if the results would be ideal. For
example, a single such peer can DoS the system by continously sending a
corrupt onion descriptor for the meta address.
Good luck with designing your P2P protocol!
More information about the tor-dev
mailing list