papers/zoo-confidential-defi/sections/09-benchmarks.tex

\section{Benchmarks}

\label{sec:benchmarks}
%----------------------------------------------------------------------

All benchmarks measured on \Zoo{} Network testnet with 100 validators, each running AMD EPYC 7763 with 256 GB RAM and NVIDIA A100 80 GB.

\subsection{Encrypted Order Operations}

\begin{table}[h]
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Operation} & \textbf{Latency (GPU)} & \textbf{Throughput} \\
\midrule
Order encryption (TFHE price + CKKS qty) & 1.2 ms & 833 orders/s \\
Price comparison (64-bit TFHE) & 4.1 ms & 243 comparisons/s \\
Volume matching (CKKS min) & 0.8 ms & 1,250 matches/s \\
Full match (compare + volume) & 4.9 ms & 204 matches/s \\
Batch match (100 orders) & 1.44 s & 340 matches/s \\
\bottomrule
\end{tabular}
\caption{Encrypted order operation benchmarks}
\label{tab:order_bench}
\end{table}

\subsection{Sealed-Bid Auction}

\begin{table}[h]
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Auction Size} & \textbf{Evaluation Time} & \textbf{Per-Bid Cost} \\
\midrule
10 bids & 41 ms & 4.1 ms \\
100 bids & 410 ms & 4.1 ms \\
1,000 bids & 4.1 s & 4.1 ms \\
10,000 bids & 41 s & 4.1 ms \\
\bottomrule
\end{tabular}
\caption{Sealed-bid auction evaluation (first-price, GPU-accelerated)}
\label{tab:auction_bench}
\end{table}

Linear scaling: each additional bid requires one encrypted comparison (4.1 ms on GPU).

\subsection{Private AMM}

\begin{table}[h]
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Operation} & \textbf{Latency} & \textbf{Notes} \\
\midrule
Encrypted swap (constant product) & 3.2 ms & CKKS arithmetic \\
Encrypted swap (concentrated) & 8.7 ms & Range check + CKKS \\
Pool state aggregation (per 100 blocks) & 124 ms & Threshold decrypt \\
LP fee calculation & 1.1 ms & CKKS multiplication \\
\bottomrule
\end{tabular}
\caption{Private AMM operation benchmarks}
\label{tab:amm_bench}
\end{table}

\subsection{Threshold Settlement}

\begin{table}[h]
\centering
\begin{tabular}{lcc}
\toprule
\textbf{Phase} & \textbf{Latency} & \textbf{Bottleneck} \\
\midrule
Match proof verification & 12 ms & Computation \\
Partial decryption (per validator) & 0.8 ms & Computation \\
Quorum collection (67-of-100) & 240 ms & Network RTT \\
Settlement transaction & 2.0 s & Block confirmation \\
\midrule
\textbf{Total (submission to settlement)} & \textbf{2.8 s} & Block time \\
\bottomrule
\end{tabular}
\caption{End-to-end settlement latency}
\label{tab:settlement_bench}
\end{table}

The 2.8-second end-to-end latency is dominated by the 2.0-second block confirmation time. Cryptographic overhead (matching + threshold decryption) adds only 0.8 seconds.

\subsection{Comparison to Plaintext DEX}

\begin{table}[h]
\centering
\begin{tabular}{lccc}
\toprule
\textbf{Metric} & \textbf{Plaintext \DEX{}} & \textbf{\Zoo{} Confidential} & \textbf{Overhead} \\
\midrule
Order submission & 0.1 ms & 1.2 ms & 12$\times$ \\
Single match & 0.01 ms & 4.9 ms & 490$\times$ \\
Batch match (100) & 0.8 ms & 1.44 s & 1,800$\times$ \\
Settlement & 2.0 s & 2.8 s & 1.4$\times$ \\
MEV extracted & \$1.5B/yr & \$0 & $-\infty$ \\
\bottomrule
\end{tabular}
\caption{Confidential DeFi overhead vs.\ plaintext execution}
\label{tab:overhead}
\end{table}

The computational overhead is significant (490$\times$ for a single match) but is amortized by batch processing and GPU acceleration. The economic overhead is negative: users save more from eliminated \MEV{} than they pay in additional gas.

%----------------------------------------------------------------------