mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-22 17:20:17 +00:00
64 lines
3.0 KiB
TeX
64 lines
3.0 KiB
TeX
\section{Consensus: Quasar}
|
|
|
|
\label{sec:consensus}
|
|
%% ============================================================================
|
|
|
|
\subsection{Overview}
|
|
|
|
Quasar is the consensus engine of the Lux Network, descended from the HotStuff family of BFT consensus protocols. It achieves finality in 420ms under normal operation, tolerates up to $f < n/3$ Byzantine validators, and supports dynamic validator set changes without halting the chain.
|
|
|
|
\subsection{Feature Matrix}
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\caption{Consensus protocol comparison.}
|
|
\footnotesize
|
|
\begin{tabular}{@{}lcccc@{}}
|
|
\toprule
|
|
Feature & \rotatebox{60}{Quasar} & \rotatebox{60}{Tendermint} & \rotatebox{60}{HotStuff} & \rotatebox{60}{Narwhal} \\
|
|
\midrule
|
|
BFT tolerance & $f < n/3$ & $f < n/3$ & $f < n/3$ & $f < n/3$ \\
|
|
Finality & 420ms & 6s & 800ms & 3s \\
|
|
TPS (observed) & 180K & 10K & 50K & 130K \\
|
|
Dynamic valset & \yes & \yes & \pmark & \yes \\
|
|
Pipelining & \yes & \no & \yes & \yes \\
|
|
PQ signatures & \yes & \no & \no & \no \\
|
|
Dual-cert & \yes & \no & \no & \no \\
|
|
Subnet support & \yes & \no & \no & \no \\
|
|
View change & $O(n)$ & $O(n^2)$ & $O(n)$ & $O(n)$ \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{table}
|
|
|
|
\subsection{Performance}
|
|
|
|
Benchmarked on a 100-node validator set deployed across 5 AWS regions (us-east-1, us-west-2, eu-west-1, ap-southeast-1, ap-northeast-1), 8-vCPU c5.2xlarge instances:
|
|
|
|
\begin{table}[H]
|
|
\centering
|
|
\caption{Consensus throughput under varying conditions.}
|
|
\footnotesize
|
|
\begin{tabular}{@{}lrrr@{}}
|
|
\toprule
|
|
Condition & TPS & Finality & CPU usage \\
|
|
\midrule
|
|
Normal (0 faulty) & 180,000 & 420ms & 62\% \\
|
|
$f = 10$ crash & 165,000 & 580ms & 71\% \\
|
|
$f = 10$ Byzantine & 142,000 & 740ms & 78\% \\
|
|
$f = 33$ crash & 98,000 & 1.2s & 85\% \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\end{table}
|
|
|
|
Tendermint's 10K TPS ceiling is well-documented in the CometBFT benchmarks. HotStuff's pipelined variant achieves 50K TPS in Meta's Diem benchmarks but has not been tested with post-quantum signatures. Narwhal+Tusk (Sui/Aptos DAG-based consensus) achieves 130K TPS in Sui's published benchmarks but with 3-second finality.
|
|
|
|
\subsection{Dual-Certificate Consensus}
|
|
|
|
Quasar is the only production consensus engine that issues dual certificates: each block is signed with both a classical Ed25519 aggregate signature and a post-quantum ML-DSA-65 (FIPS~204) aggregate signature. This provides immediate quantum resistance without sacrificing classical verification speed. Validators maintain both key pairs, and clients can verify using either signature.
|
|
|
|
\subsection{Key Differentiator}
|
|
|
|
The combination of 180K TPS, 420ms finality, and post-quantum dual-certificate signing is unique. No other consensus engine provides all three. Narwhal+Tusk approaches the throughput but at 7$\times$ higher finality. Tendermint is battle-tested (Cosmos ecosystem) but an order of magnitude slower. The subnet architecture allows application-specific consensus parameters per chain, which no alternative supports natively.
|
|
|
|
%% ============================================================================
|