Discussion: Identifying improvements for resolution systems

From a discussion that started with IPNS name resolving
@aschmahmann

go-IPFS wants IPNS to have <1 second resolution, but decentralized peerID resolution can end up taking more than 1 second. They are asking for a pseudo-centralized solution in the meanwhile. They are thinking of using something like DNS, but would it be sufficient to use rendezvous? What do you think the delivery timing is for rendezvous?

@miyazono

I’ve been kinda kicking around the idea of other, more complicated graph topologies than that of Chord in my mind and what that would look like, but that was in the context of trying overhauling the DHT for provider records, and I’m starting to wonder how connected peerID resolution is.

@aschmahmann

My thought is that PeerID resolution is halfway between a naming system resolution (e.g. IPNS) and provider records (e.g. IPFS). All three of these systems amount to advertising "You can find X at Y". IPFS provider records say "you can probably find the data with hash=Hash(data) at this address" which, even though we don't ask for confirmation they have and can provide the data (feels similar to Filecoin retrieval market issues), we know we'll get the right data when we see it. IPNS records say "You can find the data corresponding to Hash(Key) at Hash(data)", which while we can confirm with signatures we have trouble guaranteeing freshness. PeerID resolution, like IPNS, is mutable data resolution and requires freshness guarantees. However, because we can dial the peer and check that they have the correct PeerID it has the verifiable quality of IPFS provider records.

Another example of PeerIDs being pretty closely related to provider records is DHT PubSub routing. My understanding is that I advertise "I care about channel C" by putting a provider record for Hash(C) into the DHT. This effectively turns a (channel) ID into network addresses as PeerIDs do.

Start by reading the discussion on IPNS name resolving, focusing on the comparisons among rendezvous, DNS-like approaches, DHT provider records, and PeerID resolution. No files or tests are named; the work would need an agreed resolution design and clear criteria for its delivery timing and performance.