Finding Nemo-Nemo: CFT DAG-based Consensus in the WAN
A new DAG-based protocol achieves consensus in just two network hops, matching latency while significantly increasing throughput.
A team of researchers, including Rithwik Kerur, Dahlia Malkhi, and Philipp Jovanovic, has published a paper introducing Nemo-Nemo, a novel crash-fault tolerant (CFT) consensus protocol designed for high performance in wide-area networks (WAN). The protocol bridges design principles from both CFT and Byzantine-fault tolerant (BFT) worlds. Its core innovation is structuring command propagation through a causally ordered Directed Acyclic Graph (DAG), which enables a naturally self-regulating communication regime where all consensus replicas can act as proposers. This multi-leader architecture directly addresses the performance bottleneck inherent in traditional single-leader protocols like Paxos or Raft.
By cleanly separating the logic for command dissemination from the consensus logic itself, Nemo-Nemo maintains robustness even when network conditions stall consensus commits. A key feature is its handling of transient delays: leader proposals that miss a deadline are never simply dropped but are deterministically deferred and executed later, preserving system throughput. Furthermore, the protocol's design enables it to reach consensus on the DAG in just two network hops, matching the low latency of existing CFT systems while achieving significantly higher throughput. The authors position Nemo-Nemo as the first DAG-based CFT consensus protocol proven to outperform current state-of-the-art solutions in WAN environments on both speed and resilience metrics, making it a promising candidate for deployable, large-scale distributed systems.
- Uses a multi-leader DAG architecture to avoid single-leader bottlenecks, enabling all replicas to propose commands.
- Commits consensus in just two network hops, matching low latency while significantly boosting throughput over WANs.
- Handles network delays robustly by deferring, not dropping, late proposals, preserving system throughput.
Why It Matters
Enables faster, more resilient global-scale databases and blockchain systems by improving consensus over long distances.