Skip to content
Blockchain LabTHE VENTURE STUDIO

LibraryConsensus2021Design paperCorpus record

All You Need is DAG

DAG-Rider. Idit Keidar, Eleftherios Kokoris-Kogias, Oded Naor and Alexander Spiegelman.

Replicas reliably broadcast proposals into a round-structured DAG. Once the DAG is local, each replica can read a total order off it without another round of consensus messages. The paper is the asynchronous, optimally resilient version of that idea.

DAG-Rider has two layers. Reliable broadcast builds a round-by-round DAG of proposals. Each replica then reads a total order off its local DAG and sends no further consensus messages. The setting is asynchronous. The committee is known. Fewer than one third may be faulty.

The five-minute read

The DAG is the memory

A vertex names a quorum of vertices from the previous round. That quotation is how a replica proves it saw a causal history. There is no separate block header doing this job.

Order is local

Once the DAG contains the right vertices, every honest replica computes the same total order without another vote. If a protocol still runs a BFT instance per vertex, it is a different design.

Asynchrony is the point

The paper does not assume a known message-delay bound for safety or for its liveness claim. That is why the commit rule is more patient than a partially synchronous protocol.

Fairness means inclusion

Proposals from correct replicas are eventually ordered. The word is about censorship by the commit rule, not about equal fees.

One action, walked through

  1. A replica reliably broadcasts a vertex that points at a quorum of the prior round.
  2. Honest replicas eventually hold those vertices and the history they point at.
  3. The commit rule selects a wave of the DAG once enough structure is visible.
  4. The causal history of the committed vertex is ordered deterministically.
  5. A slow but correct replica is still included, which is the fairness claim.

The argument, unpacked

Zero extra messages is the claim to test

The abstract's 'all you need is DAG' means the ordering layer adds no communication. A system diagram with a DAG plus a HotStuff box has not implemented the paper.

Optimal is asymptotic

Optimal resilience, amortised communication and time are complexity claims. They are not a transactions-per-second figure. Bullshark is what you read for the common-case complaint against this paper.

What has to be true

  • A known set of replicas, with reliable authenticated channels in the model.
  • Fewer than one third Byzantine.
  • Reliable broadcast actually delivers the properties the commit rule uses.
  • Replicas store the DAG they need. The paper is not a garbage-collection design.

What happened after the paper

DAG-Rider is the citation for asynchronous DAG consensus with a local commit rule. Narwhal, Tusk and Bullshark are the follow-ons. A chain that says 'we use a DAG' still owes the commit rule and the fault assumption.

What to check before you use the idea

  • Does ordering send any messages of its own?
  • Is the network assumption asynchronous?
  • What happens to a slow honest replica's proposal?
  • Is the committee fixed in the paper's sense?

Terms

Structured DAG
A round-by-round graph in which each vertex cites a quorum of the previous round.
Local commit
A total order computed from the DAG alone, with no extra consensus round.

The problem the paper names

Classical BFT spends messages both to disseminate a proposal and to agree on its place. DAG-Rider separates those. Dissemination builds the DAG. Ordering is a local interpretation of the DAG.

What the design proposes

  • Each round, a replica broadcasts a vertex that points at a quorum of vertices from the previous round.
  • Reliable broadcast keeps the DAG consistent enough that honest replicas eventually share a causal history.
  • A commit rule walks that history. No extra votes are sent to decide the order.

How the mechanism is specified

  • Asynchrony is the network assumption. The protocol does not wait for a known bound on delay in order to stay safe.
  • The resilience claim is the usual one third. Optimal amortised communication is a complexity claim in the paper, not a throughput number for a product.
  • Fairness here means a correct replica's proposals are eventually ordered, not that users pay the same fee.

What this page does not treat as proven

  • A known committee is assumed. Open membership is outside the paper.
  • Practical garbage collection and the happy path under synchrony are what later papers, including Bullshark, add.
  • A DAG in a diagram is not this protocol unless the commit rule is the one the paper defines.

Why a venture studio still reads it

Ask whether ordering needs any messages beyond building the DAG. If the design still runs a separate PBFT on every vertex, it is not DAG-Rider.

This is Blockchain Lab's reading of a public design paper. It is not the paper, not a copy of it, and not an offer of tokens, equity, custody or a partnership. Later network behaviour can diverge from the text. Nothing here is investment, legal or technical advice.

Research status: Design paper. Last reviewed: 1 October 2026. This is a reading of a public paper, not investment, legal or security advice.