mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
74 lines
2.7 KiB
TeX
74 lines
2.7 KiB
TeX
\section{Benchmark 6: Dual-Certificate Finality}
|
|
|
|
\label{sec:dual-cert}
|
|
%% ============================================================================
|
|
|
|
This section quantifies the cost of Quasar's unique dual-certificate mechanism.
|
|
|
|
\subsection{Latency Decomposition}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrr@{}}
|
|
\toprule
|
|
\textbf{Phase} & \textbf{Duration (ms)} & \textbf{Critical Path?} \\
|
|
\midrule
|
|
Block proposal broadcast & 12 & Yes \\
|
|
Network propagation (WAN RTT) & 38--62 & Yes \\
|
|
BLS partial signing & 0.31 & No (parallel) \\
|
|
Pulsar share generation & 0.048 & No (parallel) \\
|
|
Vote + share transmission & 38--62 & Yes \\
|
|
BLS aggregation & 0.84 & Yes \\
|
|
Pulsar combination (15-of-21)& 7.0 & Yes \\
|
|
Certificate broadcast & 38--62 & Yes \\
|
|
\midrule
|
|
\textbf{Total (3 RTT + compute)} & \textbf{$\sim$\,200\,ms} & \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Finality latency decomposition for Quasar at $n=21$, 3-datacenter deployment.
|
|
The critical path is dominated by network RTT, not cryptographic computation.}
|
|
\label{tab:dual-cert-breakdown}
|
|
\end{table}
|
|
|
|
\subsection{Overhead of Dual-Certificate vs. Single-Certificate}
|
|
|
|
To isolate the cost of dual-certificate finality, we ran Quasar in two modes:
|
|
\begin{itemize}
|
|
\item \textbf{BLS-only}: Classical single-certificate (no Pulsar).
|
|
\item \textbf{Dual-cert}: Full Quasar with both BLS and Pulsar.
|
|
\end{itemize}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrrrr@{}}
|
|
\toprule
|
|
\textbf{Mode} & \textbf{Finality (ms)} & \textbf{TPS} & \textbf{Cert Size (B)} & \textbf{Overhead} \\
|
|
\midrule
|
|
BLS-only & 192 & 11{,}200 & 144 & baseline \\
|
|
Dual-cert & 204 & 10{,}400 & 1{,}168 & +6.3\% latency, $-$7.1\% TPS \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Cost of adding Pulsar post-quantum certificate to Quasar at $n=100$.}
|
|
\label{tab:dual-overhead}
|
|
\end{table}
|
|
|
|
\textbf{Key finding}: Dual-certificate finality adds only 12\,ms (6.3\%) latency and
|
|
reduces throughput by 7.1\%. The cost of quantum security is a $<$10\% performance
|
|
penalty---well within operational margin.
|
|
|
|
\subsection{Why the Overhead is Small}
|
|
|
|
Three design decisions minimize dual-certificate cost:
|
|
|
|
\begin{enumerate}
|
|
\item \textbf{Piggybacking}: Pulsar shares are sent alongside BLS partial signatures
|
|
in the same network message. No additional round trip.
|
|
\item \textbf{Threshold vs. multi-sig}: Pulsar combination is a single $\sim$7\,ms
|
|
computation, not $n$ individual verifications.
|
|
\item \textbf{Parallelism}: BLS aggregation and Pulsar combination run concurrently
|
|
on separate CPU cores. The finality deadline is
|
|
$\max(\text{BLS}, \text{Pulsar})$, not their sum.
|
|
\end{enumerate}
|
|
|
|
%% ============================================================================
|