mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
44 lines
1.5 KiB
TeX
44 lines
1.5 KiB
TeX
\section{Empirical Evaluation}
|
|
|
|
\label{sec:evaluation}
|
|
|
|
\subsection{Testnet Configuration}
|
|
|
|
We deployed Nova on a 1,000-node globally distributed testnet with parameters
|
|
$k = 20$, $\alpha = 0.8$, $\beta = 11$, DAG fan-in $= 4$, and transaction arrival
|
|
rate $T = 8{,}500$ TPS.
|
|
|
|
\subsection{Results}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrrrr@{}}
|
|
\toprule
|
|
\textbf{Byzantine \%} & \textbf{TPS} & \textbf{Median Latency} & \textbf{Avg Burst Size} & \textbf{P99 Latency} \\
|
|
\midrule
|
|
0\% & 8,500 & 300ms & 425 & 500ms \\
|
|
10\% & 8,200 & 400ms & 328 & 750ms \\
|
|
20\% & 7,800 & 500ms & 312 & 1,000ms \\
|
|
30\% & 6,500 & 700ms & 217 & 1,500ms \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Nova performance under varying Byzantine fractions}
|
|
\label{tab:nova-results}
|
|
\end{table}
|
|
|
|
Under zero Byzantine faults, Nova achieved 8,500 TPS with 300ms median latency and
|
|
average burst size of 425 transactions. This represents a 2.9x throughput improvement
|
|
over sequential finalization (which achieves $\lambda \cdot 1 \approx 2,900$ TPS with
|
|
the same consensus parameters).
|
|
|
|
\subsection{Burst Size Distribution}
|
|
|
|
The burst size distribution was approximately geometric with parameter $\lambda/T$,
|
|
consistent with Theorem~\ref{thm:burst-size}. The 95th percentile burst size was
|
|
2.3x the mean, indicating occasional ``mega-bursts'' when multiple frontier vertices
|
|
finalize in close succession.
|
|
|
|
%% -----------------------------------------------------------------------
|
|
%% 7. CONCLUSION
|
|
%% -----------------------------------------------------------------------
|