Brothers, tomorrow there will be a new-coin airdrop—make sure to set your alarm so you don’t miss it.
Last week, I used `tcpdump` on the server to capture the network packets of the @Dusk node, and I found that it barely had any Ethereum-community standard `libp2p-gossipsub` TCP handshakes. Instead, it was flooded with high-frequency UDP packets. Tracing down the Rust dependency tree into `rusk` and its underlying network library `dusk-kadcast`, I finally noticed that it directly rewrote a Kadcast broadcast protocol at the P2P transport layer.
A lot of public chains that implement ZK privacy get stuck in a physical blind spot that’s easy to overlook: **network transmission latency**. ZK-proof payloads are far larger than ordinary transfers. If you use the traditional Gossip protocol’s “random roaming” flooding, nodes create huge data redundancy and instantly choke network bandwidth. And in Dusk’s SA consensus system with 2-second block times, even a slight hiccup in proof broadcasting will trigger a verification timeout.
Kadcast’s engineering solution in `kadcast/src/peer.rs` is extremely hardcore: it fuses the Kademlia topology with **forward error correction codes (Reed-Solomon FEC)**.
When broadcasting blocks, nodes don’t send the full file. Instead, they split the ZK proof into chunks, encode it into redundant fragments, and deliver them directionally according to the Kademlia distance logic. Once the receiving node gets the required number of fragments, it can reconstruct the original proof locally within milliseconds. This directly reduces the P2P network’s data amplification factor by an order of magnitude.
Most projects think performance only depends on how fast the virtual machine (VM) runs, but they ignore that the real bottleneck for ZK throughput is the P2P network layer. Once you understand Kadcast, you realize that for Dusk to carry high-frequency financial settlement, it even rebuilt the lowest-level UDP messages at an infrastructure scale.
And for the entire Kadcast node routing scheduling and relay network, the underlying mechanics are all built on maintaining incentives and settlement#dusk $DUSK
Last week, I used `tcpdump` on the server to capture the network packets of the @Dusk node, and I found that it barely had any Ethereum-community standard `libp2p-gossipsub` TCP handshakes. Instead, it was flooded with high-frequency UDP packets. Tracing down the Rust dependency tree into `rusk` and its underlying network library `dusk-kadcast`, I finally noticed that it directly rewrote a Kadcast broadcast protocol at the P2P transport layer.
A lot of public chains that implement ZK privacy get stuck in a physical blind spot that’s easy to overlook: **network transmission latency**. ZK-proof payloads are far larger than ordinary transfers. If you use the traditional Gossip protocol’s “random roaming” flooding, nodes create huge data redundancy and instantly choke network bandwidth. And in Dusk’s SA consensus system with 2-second block times, even a slight hiccup in proof broadcasting will trigger a verification timeout.
Kadcast’s engineering solution in `kadcast/src/peer.rs` is extremely hardcore: it fuses the Kademlia topology with **forward error correction codes (Reed-Solomon FEC)**.
When broadcasting blocks, nodes don’t send the full file. Instead, they split the ZK proof into chunks, encode it into redundant fragments, and deliver them directionally according to the Kademlia distance logic. Once the receiving node gets the required number of fragments, it can reconstruct the original proof locally within milliseconds. This directly reduces the P2P network’s data amplification factor by an order of magnitude.
Most projects think performance only depends on how fast the virtual machine (VM) runs, but they ignore that the real bottleneck for ZK throughput is the P2P network layer. Once you understand Kadcast, you realize that for Dusk to carry high-frequency financial settlement, it even rebuilt the lowest-level UDP messages at an infrastructure scale.
And for the entire Kadcast node routing scheduling and relay network, the underlying mechanics are all built on maintaining incentives and settlement#dusk $DUSK
