DAG-BFT vs Chain-BFT: The Essential Architectural Split

Confidence: Certain Updated 2026-05-26 Review by 2026-09-22 Sources 6 Machine-translated Original (JA)
#systems#blockchain#BFT#consensus#mempool#architecture
On this page

Wiki route

This entry sits under systems index. Read it with fintech index for adjacent context and Smart Contract Bytecode Forensics — Three-Tier Verify Technique for the broader system boundary.

[!info] TL;DR Modern BFT consensus algorithms split into 2 major schools: Chain-BFT (HotStuff / Tendermint / Threshold Simplex / Malachite) chains blocks, while DAG-BFT (Narwhal / Bullshark / Mysticeti) promotes the mempool to a first-class consensus citizen. The essential split is not performance, but how the mempool is abstracted — Chain-BFT is suitable for settlement scenarios (audit-friendly), while DAG-BFT is suitable for high-throughput scenarios (100k+ TPS). Tempo and Arc both adopt Chain-BFT.

Comparison of the 2 major schools:

ItemChain-BFTDAG-BFT
Representative algorithmsHotStuff, Tendermint, Threshold Simplex (Tempo), Malachite (Arc)Narwhal, Bullshark, Mysticeti (Sui)
Data structureChain + view changeDAG + total order extraction
Mempool abstractionsecondary (leader gather → propose)first-class (mempool = consensus)
Time complexityO(n) per blockO(1) per block (asynchronous)
ThroughputAbout 10K TPS100K+ TPS
Finality1-2 secondssub-second
Audit / debuggingSimple (linear history)Complex (DAG traversal)
Application scenarioSettlement, stablecoins, institutional financeHigh-frequency trading, games, DEX

Why Tempo and Arc chose Chain-BFT:

  1. Audit-friendly: in settlement scenarios, auditability is required for each transaction 1 件, and unfolding DAG history is too complex
  2. TPS is already sufficient: stablecoin settlement does not require 100K+ TPS (Visa globally is about 24K TPS)
  3. High academic maturity: the papers and implementations of HotStuff and Tendermint have both been validated at large scale
  4. Formal specification verification is easier to implement: Malachite + Quint co-design is more direct under the Chain-BFT system

Scenarios where DAG-BFT is advantaged:

  • High-TPS L1s such as Sui / Aptos (contrast with Global Dex Major Five Comparison on each company’s L1 rollout)
  • Solana-like designs (PoH + Tower BFT is DAG-flavored)
  • Future AI agent micropayment scenarios (millions of sub-cent transactions per second)

Academic frontier:

  • Mysticeti (Sui, 2024): achieves sub-second finality with DAG-BFT
  • Bullshark (2022, Aptos / Sui): formalizes the first-class mempool idea
  • Threshold Simplex (2025, Tempo): engineering convergence of Chain-BFT + BLS aggregation
  • Malachite (2025, Arc): Chain-BFT + Quint formal verification paradigm

Paper index:

  • HotStuff: Yin et al. 2019 (https://arxiv.org/abs/1803.05069)

  • Narwhal & Bullshark: Spiegelman et al. 2022

  • Mysticeti: Babel et al. 2024

  • Threshold Simplex: Tempo blog / Paradigm research

  • Malachite: Informal Systems 2025

Application decision framework:

  • Settlement / clearing → Chain-BFT (contrast with institutional chains’ preference for linear auditability in protocol renewal trigger)
  • Games / high frequency → DAG-BFT
  • General-purpose L1 → either can work (depending on ecosystem goals)

Sources