mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
58 lines
2.3 KiB
TeX
58 lines
2.3 KiB
TeX
\section{Benchmark 4: Message Complexity}
|
|
|
|
\label{sec:messages}
|
|
%% ============================================================================
|
|
|
|
We measure actual network messages exchanged per committed block, not asymptotic
|
|
complexity class.
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\begin{tabular}{@{}llrr@{}}
|
|
\toprule
|
|
\textbf{Protocol} & \textbf{Complexity Class} & \textbf{Msgs @ $n=21$} & \textbf{Msgs @ $n=100$} \\
|
|
\midrule
|
|
\textbf{Quasar} & O($n$) & 44 & 202 \\
|
|
HotStuff & O($n$) & 63 & 300 \\
|
|
Snow & O($K \cdot \beta$) & 500 & 500 \\
|
|
Narwhal+Tusk & O($n$) & 84 & 400 \\
|
|
Tower BFT & O($n$) & 42 & 200 \\
|
|
Tendermint & O($n^2$) & 441 & 10{,}000 \\
|
|
Casper FFG & O($n$) & 42 & 200 \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Messages per decision round. Snow's count is constant ($K \cdot \beta$) but high
|
|
per round; it requires multiple rounds for finality.}
|
|
\label{tab:messages}
|
|
\end{table}
|
|
|
|
\subsection{Quasar Message Breakdown}
|
|
|
|
Per committed block, Quasar exchanges:
|
|
\begin{itemize}
|
|
\item 1 block proposal broadcast (leader $\to$ all): $n$ messages
|
|
\item $n$ BLS partial signature responses (validators $\to$ leader): $n$ messages
|
|
\item 1 BLS aggregate broadcast (leader $\to$ all): integrated with next proposal
|
|
\item Pulsar shares collected in parallel: $n$ messages (piggybacked on BLS response)
|
|
\item 1 dual-certificate broadcast: integrated with next proposal
|
|
\end{itemize}
|
|
|
|
Total: $\approx 2n$ messages per block in steady state, yielding O($n$) complexity.
|
|
This is achieved by piggybacking Pulsar shares on the BLS vote message, avoiding a
|
|
separate collection round.
|
|
|
|
\subsection{Comparison with PBFT-family}
|
|
|
|
Tendermint inherits PBFT's all-to-all communication in the Prevote and
|
|
Precommit phases~\cite{tendermint}.
|
|
At $n{=}100$, each phase produces $100^2{=}10{,}000$ messages.
|
|
At $n{=}1{,}000$ this grows to $10^6$ messages per phase---the root
|
|
cause of Tendermint's scalability cliff (Section~\ref{sec:scalability}).
|
|
|
|
HotStuff~\cite{hotstuff} reduces this to O($n$) by funneling all votes through
|
|
the leader, which Quasar inherits. Quasar adds only a constant overhead
|
|
(${\sim}1.4{\times}$ HotStuff) for the Pulsar shares, since they are
|
|
piggybacked rather than sent separately.
|
|
|
|
%% ============================================================================
|