#dusk @Dusk went down a networking rabbit hole tonight instead of the usual application layer stuff, specifically why dusk bothers replacing plain gossip with something called kadcast, and honestly? 😂 i didnt expect networking topology to actually matter this much for consensus performance.
so most blockchains just flood messages to all connected peers, which is simple but wasteful, lots of redundant transmissions. kadcast instead builds a structured overlay based on kademlia, every peer gets a 128 bit id from hashing its ip and port, and messages get forwarded along deterministic paths based on xor distance between peer ids rather than blasted everywhere at once. peers keep k buckets of known neighbors, and a small delegate count per bucket handles onward broadcasting instead of every single peer relaying to every other peer. the measured result is somewhere around 25 to 50 percent lower bandwidth usage compared to gossip, with more predictable latency instead of gossips unpredictable fan out.
what i keep coming back to since digging into this is a concern thats pretty well known in kademlia research generally, structured routing based on deterministic id distance means positioning in that id space isnt random. i havent seen dusk specific writing addressing whether that opens any eclipse or targeted positioning risk around the committee vote propagation i wrote about with succinct attestation a while back, versus a fully unstructured gossip network where that kind of positioning wouldnt really be possible in the same way.
still, a real measured bandwidth reduction for consensus messages is a genuinely concrete efficiency gain, not just a cleaner sounding architecture on paper.
honestly dont know if that structured routing tradeoff introduces meaningful new risk or if the efficiency gain clearly wins out here??
🤔
$DUSK