papers/lp-010-quasar-stm-4/sections/06-lane-class-nemo.tex

\section{LaneClass (NEMO)}

\label{sec:laneclass}
%% ============================================================================

\subsection{Classes}

NEMO~\cite{nemo2025} observes that, post-consensus, the bottleneck is
execution and that blockchain workloads have heavy access skew. The
prescription is to classify state by ownership and contention pattern,
and to drive validation policy from the class:

\begin{lstlisting}[language=C++]
enum class LaneClass : uint8_t {
    Owned       = 0,   // account-local / nonce-local / private state
    Shared      = 1,   // shared contract state, ordinary contention
    HotShared   = 2,   // AMM reserves, order-book level, fee counter
    Commutative = 3,   // additive / append / reducer lane
    Unknown     = 4,   // fall through to ordinary speculative path
};
\end{lstlisting}

\subsection{Per-class policy}

Table~\ref{tab:laneclass-policy} shows the production policy mapping.

\begin{table}[ht]
\centering
\small
\caption{LaneClass execution and validation policy (4.0 production).}
\label{tab:laneclass-policy}
\begin{tabular}{p{0.18\linewidth}p{0.40\linewidth}p{0.32\linewidth}}
\toprule
\textbf{Class} & \textbf{Validation policy} & \textbf{Versioning mode} \\
\midrule
\texttt{Owned}       & no validation against unrelated txs; fast commit if owner / nonce match               & \texttt{SingleVersionFast} \\
\texttt{Shared}      & ordered MVCC (Tier~2)                                                                  & \texttt{MultiVersion} \\
\texttt{HotShared}   & semantic reducer / serialized lane / precompile path                                  & \texttt{Reducer} or \texttt{Serialized} \\
\texttt{Commutative} & Tier~3 reducer commit (\S\ref{sec:tier3})                                              & \texttt{Reducer} \\
\texttt{Unknown}     & ordinary speculative Block-STM (Tier~1 + Tier~2) per 3.0~\cite{lp010-quasarstm-3}     & \texttt{MultiVersion} \\
\bottomrule
\end{tabular}
\end{table}

\subsection{Classification sources}

Per round, every lane has a class. Classes are derived from:
\begin{enumerate}[leftmargin=2em]
\item \textbf{Static ABI annotations} on contracts (e.g. a Solidity
      \texttt{lane\_class("Commutative")} attribute on a state
      variable).
\item \textbf{Precompile self-declaration}: precompiles declare the
      lane class of every state variable they touch. The DEX matching
      precompile, for example, declares its order book level lanes
      \texttt{HotShared}.
\item \textbf{Hot-lane telemetry promotion}: a lane that exceeds
      \texttt{HOT\_LANE\_CONFLICT\_THRESHOLD} mismatches in round $n$
      is promoted to \texttt{HotShared} for round $n+1$
      (\S\ref{sec:tier1}).
\item \textbf{Predictor}~\cite{foresight2025}: classified by
      observed-conflict-rate quantile.
\item \textbf{Default}: \texttt{Unknown}---falls through to speculative
      Block-STM, exactly as in 3.0.
\end{enumerate}

Classification is finalised at the start of the round and frozen for
the round's duration.

\subsection{Why this matters for throughput}

NEMO's central observation is that the throughput ceiling is bounded by
the share of \texttt{Owned} and \texttt{Commutative} traffic. Skewed
regulated-DEX workloads (per-account orders, fee accumulation, balance
deltas) have a heavy \texttt{Owned} and \texttt{Commutative} tail, so
Tier~1 fast path and Tier~3 reducer commit each absorb a large
fraction of the load:

\begin{itemize}[leftmargin=2em]
\item \texttt{Owned}: Tier~1 passes always (no cross-tx validation)
      $\Rightarrow$ commits in O$(1)$.
\item \texttt{Commutative}: Tier~3 reducer commit at horizon
      $\Rightarrow$ no MVCC contention.
\item \texttt{HotShared}: routed to a serialized lane or to a
      reducer-mode lane $\Rightarrow$ no speculative repair storm.
\item \texttt{Shared} and \texttt{Unknown}: ordinary Block-STM
      $\Rightarrow$ inherits 3.0 behaviour~\cite{lp010-quasarstm-3}.
\end{itemize}

%% ============================================================================