mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
98343d238e
Quasar Horizon (consensus composition over Pulsar / ML-DSA / BLS lanes
under Prism + LSS lifecycle), Warp 2.0 (Pulse-backed cross-chain envelope
with forward-only v1->v2 migration and lattigo ReadUint64Slice DoS fix),
benchmark report (microbench, reshare, activation, Warp 2.0 verify, Prism
composition, GPU headroom, cross-WAN latency), and NIST IR 8214C-style
submission package (spec, parameter sets, domain separation, serialization,
reference-impl pointers, evaluation report, security analysis, side-channel
considerations).
All four deliverables compile cleanly via pdflatex+bibtex and reference
the canonical proof bucket files at proofs/{definitions,pulsar,lss,quasar}.
No duplication of proof bodies; cite by file path.
82 lines
2.4 KiB
TeX
82 lines
2.4 KiB
TeX
\section{Empirical Evaluation}
|
|
|
|
\label{sec:evaluation}
|
|
|
|
\subsection{Gate Performance}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrr@{}}
|
|
\toprule
|
|
\textbf{Gate} & \textbf{Time (ms)} & \textbf{Ciphertext Size (bytes)} \\
|
|
\midrule
|
|
NOT & 0.001 & 8,208 \\
|
|
AND & 12.3 & 8,208 \\
|
|
OR & 12.1 & 8,208 \\
|
|
XOR & 12.4 & 8,208 \\
|
|
NAND & 12.2 & 8,208 \\
|
|
NOR & 12.3 & 8,208 \\
|
|
XNOR & 12.5 & 8,208 \\
|
|
MUX & 24.6 & 8,208 \\
|
|
MAJORITY & 12.4 & 8,208 \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{TFHE gate evaluation performance (single core)}
|
|
\label{tab:gate-perf}
|
|
\end{table}
|
|
|
|
\subsection{Multi-Bit Operations}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrr@{}}
|
|
\toprule
|
|
\textbf{Operation} & \textbf{Bits} & \textbf{Time (ms)} \\
|
|
\midrule
|
|
Addition (uint32) & 32 & 780 \\
|
|
Comparison (uint32) & 32 & 395 (CMPCOMBINE) \\
|
|
Multiplication (uint8) & 8 & 1,200 \\
|
|
Equality (uint32) & 32 & 420 \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Multi-bit operation performance}
|
|
\label{tab:multibit}
|
|
\end{table}
|
|
|
|
\subsection{Threshold Decryption}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}lrr@{}}
|
|
\toprule
|
|
\textbf{Threshold} & \textbf{Decryption Time} & \textbf{Communication} \\
|
|
\midrule
|
|
2-of-3 & 5ms & 24 KB \\
|
|
3-of-5 & 12ms & 40 KB \\
|
|
5-of-10 & 28ms & 80 KB \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Threshold decryption performance}
|
|
\label{tab:threshold-perf}
|
|
\end{table}
|
|
|
|
\subsection{F-Chain NTT primitive on Apple Metal}
|
|
|
|
The F-Chain TFHE pipeline reduces every gate evaluation to NTT-domain
|
|
multiplications. The Metal NTT kernel (\texttt{luxcpp/fhe}) reaches
|
|
${\sim}23.63\times$ CPU at $N{=}4{,}096$, $B{=}128$ on Apple M1 Max
|
|
(LP-137 Phase-2 roll-up, 2026-04-26 reproduction; Phase-3 unchanged).
|
|
Smaller batches degrade gracefully: $9.0\times$ at $N{=}4{,}096$,
|
|
$B{=}32$, and $6.2\times$ at $N{=}8{,}192$, $B{=}128$. This is one
|
|
of three production workloads on the LP-137 substrate that beat CPU
|
|
end-to-end (alongside C-Chain and B-Chain BLS aggregate batched
|
|
verify) and is plausibly state-of-the-art on Apple Silicon for the
|
|
TFHE NTT shape. The full empirical surface (CPU, Metal, MLX, CUDA,
|
|
WGSL) is exercised by 2 174 gtest cases across the four backends
|
|
that compile on each test host; CPU is the byte-equivalence ground
|
|
truth that every backend is asserted against.
|
|
|
|
%% -----------------------------------------------------------------------
|
|
%% 8. CONCLUSION
|
|
%% -----------------------------------------------------------------------
|