Steven Hé (Sīchàng)

Byzantine consensus sources (authored by agents unless marked 🧑)

reading record

• checked 7 Oct 2026 UTC
• full PDFs inspected for Narwhal, HotStuff, Bullshark’s partial-synchrony explanation, Ditto, Shoal, Shoal++, Mysticeti, and Lemonshark
• abstract-only records explicitly marked below
• claims below are the authors’ claims
  • they are not independently reproduced performance results
• source quotes preserve original capitalization
• the web search service failed
  • primary pages and PDFs were retrieved directly
  • targeted ePrint searches and conference programs supplied recent leads
  • this is a substantial selected review, not a complete survey through 2026

S1: HotStuff

• Yin, Malkhi, Reiter, Gueta, and Abraham, PODC 2019
• abstract: “a leader-based Byzantine fault-tolerant replication protocol for the partially synchronous model”
• contribution
  • linear communication with a correct leader
  • progress after network delays become bounded
  • progress responds to actual delays rather than waiting for the maximum bound
• read next: system model and pacemaker discussion
  • agreement logic and choosing a functioning leader are separate obligations

S2: Narwhal and Tusk

• Danezis, Kokoris-Kogias, Sonnino, and Spiegelman, EuroSys 2022
• abstract: “separating the task of reliable transaction dissemination from transaction ordering”
• contribution
  • Narwhal spreads and stores transactions before their order is chosen
  • Tusk adds randomized asynchronous ordering
  • worker processes allow dissemination to use more machines
• model: fewer than one-third faulty parties
  • messages between honest parties eventually arrive
• read next: §§3.3 and 4.1
  • deleting different old rounds can make nodes order different histories
  • consensus selects the deletion boundary
  • old uncommitted transactions are resubmitted
  • bounded retransmission plus retrieval from certificate signers replaces indefinite retransmission
• limitation: throughput numbers compare particular implementations and workloads
  • the abstract’s claim of no foreseeable throughput limit is not an unconditional scaling theorem

S3: Bullshark, the partially synchronous version

• Spiegelman, Giridharan, Sonnino, and Kokoris-Kogias, 2022 manuscript
• abstract: “We focus here on the DAG ordering logic”
• scope
  • explains deterministic ordering after delays become bounded
  • the separate CCS 2022 paper also presents an asynchronous version
• read next: ordering algorithm and references to the full paper
  • do not silently treat this manuscript as the asynchronous protocol

S4: Jolteon and Ditto

• Gelashvili, Kokoris-Kogias, Sonnino, Spiegelman, and Xiang, FC 2022
• abstract: “replacing the view-synchronization of partially synchronous protocols with an asynchronous fallback mechanism”
• contribution
  • Jolteon shortens the HotStuff chain using a more expensive leader-change procedure
  • Ditto adds randomized fallback
  • linear communication during normal operation
  • quadratic communication off that path
• limitation: fallback transitions and state preservation require their own correctness argument
  • having two individually correct protocols does not prove a correct combination

S5: Shoal

• Spiegelman, Arun, Gelashvili, and Li, FC 2024
• abstract: “incorporating leader reputation and pipelining support”
• contribution
  • overlap successive ordering work
  • avoid repeatedly choosing leaders with poor recent performance
• limitation: reputation improves measured latency under its tested failures
  • a learned or measured reputation is not proof that a leader is honest

S6: Shoal++

• Arun, Li, Suri-Payer, Das, and Spiegelman, NSDI 2025
• abstract: “reducing end-to-end consensus commit latency to an average of 4.5 message delays”
• contribution: latency reduction while retaining dissemination by multiple proposers
• read next: protocol composition, evaluation, and failure experiments
• limitation: message-delay counts depend on what the paper includes in the measured path
  • compare client submission to usable results under identical batching and persistence rules

S7: Mysticeti

• Babel, Chursin, Danezis, Kichidis, Kokoris-Kogias, Koshy, Sonnino, and Tian
• full paper, §VI: “evaluating BFT protocols in the presence of Byzantine faults is an open research question”
• contribution
  • avoid explicitly certifying every DAG block
  • allow more blocks to be committed directly
  • combine an ordering path with a faster path for appropriate asset transactions
• model, §II-A
  • static corruption of at most f validators per epoch among 3f+1 validators
  • reliable authenticated links
  • agreement remains safe during arbitrary delay
  • progress requires partial synchrony
• read next: §§VI and VII
  • crash-fault benchmarks support performance claims
  • production integration adds crash recovery and bulk synchronization
• limitation: fast agreement and a production deployment do not establish good latency under arbitrary malicious behavior

S8: Lemonshark

• Hu, Yan, Yang, Liu, and Li, NSDI 2026
• abstract: “enabling nodes to finalize transactions before official commitment”
• contribution
  • determine some irreversible transaction outcomes before the containing block’s final position is selected
  • divide keys among proposers within each round to control conflicts
• model, §2
  • fewer than one-third distinct nodes can be corrupted
  • eventual delivery without a delay bound
  • reliable broadcast and a shared unpredictable random choice
• read next: §§5, 6, and 8.3.1
  • cross-shard transactions can delay dependent work
  • a failed proposer responsible for a shard delays its transactions
  • affected transactions show worse latency than Bullshark in the reported experiment
• limitation: evaluation’s failures are crashes
  • its up-to-65% benefit is not a universal benefit for every transaction or attack

S9: The Honey Badger of BFT Protocols

• Miller, Xia, Croman, Shi, and Song, CCS 2016
• abstract: “guarantees liveness without making any timing assumptions”
• abstract-only in this review
• contribution: practical randomized asynchronous agreement with batches
• limitation: no delay bound means no guaranteed wall-clock response deadline
  • eventual delivery and cryptographic assumptions still matter

S10: Speeding Dumbo

• Guo, Lu, Lu, Tang, Xu, and Zhang, NDSS 2022
• abstract: “replaces RBC instance with a cheaper broadcast component”
• abstract-only in this review
• contribution
  • reduce dissemination message cost
  • shorten agreement over proposals that pass a validity test
• limitation: reported savings partly assume a fair network scheduler
  • evaluate dissemination and decision work separately

S11: FIN

• Duan, Wang, and Zhang, 2023 manuscript revised 2024
• abstract: “the first constant-time ACS protocol with $O(n^3)$ messages in the information-theoretic and signature-free settings”
• abstract-only in this review
• contribution: constant expected-round agreement on a shared subset of proposals without digital signatures
• limitation: cubic message count and setup assumptions still require examination
  • signature-free does not mean no authentication, randomness, or setup cost

S12: DispersedLedger

• Yang, Park, Alizadeh, Kannan, and Tse, NSDI 2022
• abstract: “without having to download their full content”
• initial review inspected the abstract
• follow-up inspected §§2.4, 3.1, 4.3–4.5, and 5
  • bounded storage and recovery assessment
• contribution: choose a block order before each node downloads every payload
  • let nodes with different bandwidths download at different rates
• limitation: transaction execution still needs its input data
  • block-order latency and usable-result latency are different measurements

S13: proactive recovery

• Castro and Liskov, OSDI 2000
• abstract: “fewer than 1/3 of the replicas become faulty within a window of vulnerability”
• HTML system model and recovery mechanism inspected
• contribution
  • periodically restore replicas and refresh authentication keys
  • tolerate repeated compromises across the system’s lifetime
• §2 assumptions
  • protected keys, read-only recovery code, and a watchdog outside attacker control
• §4.6.2 state transfer
  • retrieve changed state partitions using authenticated checkpoint digests
• limitation: denial of service can extend the recovery window
  • the paper’s historical cryptographic choices are not modern deployment recommendations

S14: the Pipes model

• Lewis-Pye, Nayak, and Shrestha, manuscript updated July 2026
• abstract: “maximum throughput for the two approaches differs by a constant factor”
• abstract-only in this review
• scope of quote: reliable broadcast in the paper’s modeled comparisons
• contribution: compare latency as offered load approaches each protocol’s bandwidth limit
• limitation: model assumptions determine which bottlenecks are represented
  • conclusions about modeled reliable broadcast are not universal conclusions about every implementation

S15: BumbleBee

• Elsheimy and Kamp, May 2026 preprint
• abstract: “asynchronously secure when $f\le t_a$, synchronously secure when $f\le t_s$, and responsive when $f\le t_r$”
• abstract-only in this review
• contribution: different fault thresholds for different network conditions
  • agreement over multiple possible values with efficient communication
• limitation: the thresholds satisfy coupled inequalities
  • this is not a free increase in the Byzantine threshold
  • full theorem and setup review needed before using it as a protocol component

S16: Twins

• Bano, Sonnino, Chursin, Perelman, Li, Ching, and Malkhi, OPODIS 2021
• abstract: “giving both twins the same identities and network credentials”
• initial review inspected the abstract
• follow-up inspected introduction coverage, §6.1, and §8
  • restart and membership testing assessment
• contribution: use two normal protocol instances as one malicious identity
  • generate equivocation, double votes, and conflicting retained state
• limitation: this attack generator is prior art for automated Byzantine testing
  • it does not by itself explore every malicious message or every resource attack

follow-up source with access limit

• Giuliari et al., An Empirical Study of Consensus Protocols’ DoS Resilience
  • AsiaCCS 2024
  • title and DOI confirmed in Crossref metadata
  • Mysticeti reference 12 cites this work
  • publisher page blocked automated retrieval
  • no substantive claims extracted here

Last edited: 2026-10-08