DAG-Based BFT Protocols
Stub. A recent, distinct architectural approach to BFT consensus: separate data dissemination (building a directed acyclic graph of referenced blocks from all parties concurrently) from ordering (a lightweight, infrequent consensus step over the DAG's structure), which decouples throughput from per-round leader bottlenecks. Covers the "DAG meets BFT" framing, Sailfish and Shoal++ as latency/throughput-optimized instances, graded dispersal with perfect security as a building block, and Mysticeti as a deployed example.
Classical leader-based BFT protocols (the HotStuff/Simplex families elsewhere in this topic) bottleneck throughput on a single leader disseminating data each round. DAG-based BFT protocols instead let every party continuously broadcast blocks that reference prior blocks from other parties, building a directed acyclic graph of causally-related data entirely independent of any ordering decision; a separate, much cheaper consensus mechanism then runs on top of the DAG’s structure to extract a total order — reusing asynchronous-agreement-protocols‘s building blocks (reliable broadcast, binary agreement) for the ordering step while dissemination proceeds at full network speed. Covers this general “DAG meets BFT” architecture, Sailfish (improving DAG-BFT latency specifically) and Shoal++ (pushing for high throughput without sacrificing speed), graded dispersal with perfect security as a dissemination-layer building block, and Mysticeti as a concrete deployed instance (used in the Sui blockchain).
Outline (TODO — flesh out each)
- DAG meets BFT — separating dissemination from ordering
- Sailfish — improving the latency of DAG-based BFT
- Shoal++ — high throughput without sacrificing speed
- What DAGs buy BFT consensus, structurally
- Graded dispersal with perfect security
- Mysticeti