Source linked

FinWhale Cuts DAG Consensus to Two Rounds, Matches Optimal Resilience

arxiv.org@proud_deer3 hours ago·Systems Engineering·0 comments

By combining a novel fast path commit with Mysticeti's slow path, FinWhale becomes the first DAG-based BFT protocol to achieve two-message-delay termination under partial synchrony.

finwhalemysticetibyzantine fault tolerancedag consensusdistributed systemsconsensus protocols

Existing DAG-based Byzantine Fault Tolerant protocols need at least three message delays to commit a decision. FinWhale cuts that to two.

Two-Round Fast Path, No Sacrifice

FinWhale extends the Mysticeti DAG protocol with a fast path commit mechanism that coexists with the original slow path rules. The key trick: new commit structures called fast path evidence blocks. These let validators combine fast path and slow path reasoning consistently across different local DAG views. No conflicts, no safety violations.

Prior fast path BFT protocols achieved two-message-delay termination but didn't extend to DAGs. DAGs gave throughput at the cost of latency. FinWhale shows you can have both.

The Math Behind the Resilience

FinWhale operates in the partially synchronous model with $n = 3f + 2p - 1$ validators. That matches the known lower bound for fast Byzantine consensus. The protocol tolerates up to $f$ Byzantine faults. During the fast path, it achieves fast termination whenever at most $p$ validators fail, where $p$ is between $1$ and $f$. A flexible parameter that lets operators tune for expected fault patterns.

Why the extra term $2p - 1$? It accounts for the fast path's need for a quorum of honest validators that see the same fast evidence. The paper's authors prove the bound is tight.

How FinWhale Avoids Safety Collisions

DAG protocols rely on partial orders of blocks. A fast path commit that happens in two rounds must not override a slow path commit from three rounds later. FinWhale's evidence blocks act as cryptographic anchors that tie fast path decisions to the DAG structure. Validators check consistency before accepting a fast path commit. If conditions degrade, they fall back to Mysticeti's three-round slow path with no interruption.

The analysis covers all Byzantine adversaries that can equivocate, delay messages, or corrupt validators up to threshold $f$. The safety proof is in the paper; I won't reproduce it here.

FinWhale proves that optimal-latency fast paths are practical in uncertified DAG consensus, opening the door for even lower-latency blockchain and distributed ledger designs.

Source: FinWhale: An Optimally Resilient Two-Round Terminating DAG Protocol
Domain: arxiv.org

Read original source ->

External source stays available while the OJO article and comment thread stay local.

More in Systems Engineering

view topic
Thermal-Load Balancing: Stop Orbital AI Servers From Cooking Themselves

LLM training in space demands sub-10μs latency, packing GPUs so tight that heat crosstalk throttles throughput and melts hardware before it can amortize rocket launch carbon.

AROM Flips the OS-DIMM Contract to Fix LtRAM's Slow Writes

Instead of hiding LtRAM's asymmetric write latency behind a hardware translation layer like Intel Optane, AROM makes application memory read-only and lets the OS handle writes via copy-on-write page migrations.

Hash-Based Address Translation Cuts TLB Miss Latency 15% Over Prior Art

Revelator, a hardware-OS cooperative technique from CMU-SAFARI, uses hashing to predict physical addresses before translation completes, achieving 1.50x speedup over huge pages under high fragmentation.

Calibrating Query Costs for Confidential VMs Recovers 48% Performance

A lightweight cost calibration for AMD SEV-SNP confidential VMs narrows the performance gap with KVM, recovering up to 48% and even beating the baseline on some workloads.

Comments load interactively on the live page.