Duet paper boosts blockchain throughput 7.26x by co-optimizing P2P and consensus

In blockchain systems, peer-to-peer (P2P) overlay networks are critical for reliable message delivery, but existing designs treat the consensus layer and P2P network as separate opaque components, missing optimization opportunities. A new paper titled "Duet: Co-Optimizing P2P Message Propagation and Rotating-Leader Consensus" proposes a co-design approach that exploits blockchain's state machine abstraction to record network topologies and latencies in a trusted, coordinated way. The authors—Yifeng Ye, Rongji Huang, Gerui Wang, Mingchao Wan, Yuxing Duan, Jingjing Zhang, and Shengyun Liu—present three key improvements: accelerating leader rotation using network state information; introducing a reliable broadcast paradigm that uses tree-based dissemination during normal operation with gossip fallback only when needed; and constructing latency-aware dissemination trees that minimize propagation delays.

To validate the approach, the team integrated Duet into Tendermint (a popular Byzantine fault-tolerant consensus protocol) and libp2p (a modular P2P networking stack). They ran large-scale experiments on Amazon EC2 with up to 300 nodes distributed across 10 geographic regions. Results show that compared to standard gossip-based dissemination over the same topology, Duet improves peak throughput by up to 7.26x. This breakthrough demonstrates that co-optimizing the P2P and consensus layers can dramatically enhance blockchain performance without requiring new hardware or changing the underlying protocol security properties. The work has implications for layer-1 blockchains, distributed ledgers, and any state machine replication system where latency and throughput are critical.