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.
106 lines
4.3 KiB
TeX
106 lines
4.3 KiB
TeX
\section{Commit horizon}
|
|
|
|
\label{sec:horizon}
|
|
%% ============================================================================
|
|
|
|
\subsection{Horizon as a finalisation primitive}
|
|
|
|
QuasarSTM~3.0 sketched the commit horizon: a contiguous Nova prefix or
|
|
a valid Nebula causal cut that can be finalised in one batch instead of
|
|
transaction-by-transaction~\cite{lp010-quasarstm-3}. 4.0 ships the
|
|
production semantics.
|
|
|
|
\begin{lstlisting}[language=C++]
|
|
struct CommitHorizon {
|
|
uint64_t prefix_len; // Nova: number of canonical-order txs committed
|
|
Hash causal_cut_root; // Nebula: commitment over the cut frontier
|
|
Hash reducer_state_root; // post-reduce snapshot for all Reducer lanes
|
|
uint64_t round_index; // Quasar round this horizon belongs to
|
|
uint64_t advanced_at_tick; // wave-tick index where this horizon advanced
|
|
uint8_t mode; // 0=Nova, 1=Nebula
|
|
uint8_t pad[7];
|
|
};
|
|
\end{lstlisting}
|
|
|
|
\subsection{Advance condition}
|
|
|
|
A horizon advance over canonical-order range $[t_a, t_b]$ requires:
|
|
\begin{enumerate}[leftmargin=2em]
|
|
\item Every $t \in [t_a, t_b]$ has passed Tier~1 / Tier~2 / Tier~3 (as
|
|
applicable).
|
|
\item No repair is outstanding for any $t$ in the range.
|
|
\item Every \texttt{Reducer}-mode lane touched in the range has run
|
|
\texttt{deterministic\_reduce\_at\_commit} (\S\ref{sec:tier3}).
|
|
\item For Nebula, the cut frontier is a valid antichain in the DAG.
|
|
\end{enumerate}
|
|
|
|
When the four conditions hold, the horizon advances to $t_b$, the
|
|
roots are computed once, and the per-tx receipts are emitted. Per-tx
|
|
roots are not emitted; this is what slashes per-transaction certificate
|
|
overhead.
|
|
|
|
\subsection{Roots emitted at horizon advance}
|
|
|
|
At each horizon advance, the round-result roots
|
|
(\texttt{block\_hash}, \texttt{state\_root}, \texttt{receipts\_root},
|
|
\texttt{execution\_root}, \texttt{mode\_root}) are recomputed and the
|
|
new \texttt{hint\_roots}~\cite{chiron2025} are appended:
|
|
|
|
\[
|
|
\mathit{hint\_roots} = \langle
|
|
\mathit{access\_spec\_root},\;
|
|
\mathit{dependency\_graph\_root},\;
|
|
\mathit{hot\_lane\_root},\;
|
|
\mathit{conflict\_profile\_root},\;
|
|
\mathit{reducer\_plan\_root}
|
|
\rangle
|
|
\]
|
|
|
|
The \texttt{hint\_roots} ride in the existing \texttt{roots} slot of
|
|
\texttt{QuasarRoundResult}---no wire-format change. They are
|
|
non-load-bearing: replay validators consume them to schedule faster,
|
|
but safety does not depend on them~\cite{chiron2025}.
|
|
|
|
\subsection{Horizon safety}
|
|
|
|
\begin{theorem}[Horizon safety]
|
|
\label{thm:horizon-safety}
|
|
For any commit horizon $H$ advanced from $t_{a-1}$ to $t_b$ in a Quasar
|
|
round, the post-horizon state is canonically equivalent to the
|
|
sequential execution of $t_a, t_{a+1}, \ldots, t_b$ in canonical order,
|
|
modulo the reduction rule for \texttt{Reducer}-mode lanes
|
|
(\S\ref{sec:tier3}).
|
|
\end{theorem}
|
|
\begin{proof}[Proof sketch]
|
|
For $t \in [t_a, t_b]$ in \texttt{SingleVersionFast} or
|
|
\texttt{MultiVersion}, Tier~1 and Tier~2 validation guarantee that
|
|
$t$'s reads see the canonical visible version (3.0 \S\,5,
|
|
\cite{lp010-quasarstm-3,adya2000essn}). For \texttt{Reducer}-mode
|
|
lanes, Algorithm~\ref{alg:reduce} applies operations in canonical
|
|
order over \texttt{tx\_id}, which is the same order as sequential
|
|
execution. \texttt{Serialized} mode trivially executes writes in
|
|
canonical order. Hence the post-horizon state matches a sequential
|
|
execution that handles \texttt{Reducer} lanes by applying their
|
|
operations in tx-id order.
|
|
\end{proof}
|
|
|
|
\subsection{Why per-round, not per-tick}
|
|
|
|
A round may advance the horizon multiple times within its wave-tick
|
|
budget. The horizon is per-round; each advance emits one set of roots.
|
|
The certificate (LP-020) is emitted at \emph{round} end and binds
|
|
\emph{the final horizon} of the round, not intermediate horizons.
|
|
Intermediate horizons are convenient checkpoints for replay validators
|
|
but are not certified.
|
|
|
|
\begin{remark}[Nova vs.\ Nebula horizon]
|
|
In Nova mode, the horizon is always a prefix of the canonical order
|
|
($t_b - t_{a-1}$ contiguous transactions). In Nebula mode, the horizon
|
|
is a causal cut (an antichain in the DAG); the canonical order is
|
|
extended by the deterministic tie-break rule from
|
|
3.0~\cite{lp010-quasarstm-3} so that reduction order over
|
|
\texttt{tx\_id} remains well-defined.
|
|
\end{remark}
|
|
|
|
%% ============================================================================
|