mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
dce991ae1d
The seven engine specifications (lux-{photon,wave,nova,nebula,prism,ray,field}-protocol)
were fragmented short notes documenting layers of the same Quasar consensus stack.
This commit merges them into a single canonical reference, lux-consensus-engines.tex,
that uses one preamble, one set of theorems, one bibliography, and one composition
chapter showing how the engines layer into the LP-020 finality pipeline.
The seven originals are reduced to single-page breadcrumb stubs pointing at the
consolidated paper; their sections/ content is preserved on disk for git history.
- new: lux-consensus-engines/ (50 pages, 646KB PDF, ~2900 lines tex)
- breadcrumbs: each original keeps cover + title + stub paragraph naming its chapter
- composition: chapter 9 records the layered contract Photon < Wave < Nova < ...
- bibliography: consolidated from the seven original thebibliography blocks
35 lines
2.0 KiB
TeX
35 lines
2.0 KiB
TeX
\section{Introduction}
|
|
|
|
\label{sec:photon-introduction}
|
|
|
|
In standard metastable consensus, a validator processes one transaction at a time per
|
|
conflict set: it samples, waits for responses, updates confidence, and repeats for
|
|
$\beta$ rounds. This creates a throughput ceiling of $1/(\beta \cdot \Delta)$ finalizations
|
|
per second per conflict set, where $\Delta$ is the round time. For $\beta = 11$ and
|
|
$\Delta = 50$ms, this is merely 1.8 finalizations per second per conflict set.
|
|
|
|
While the Nova protocol amortizes this cost across DAG subgraphs, independent conflict
|
|
sets still bottleneck at the per-conflict finalization rate. Photon addresses this by
|
|
pipelining: multiple transactions enter the sampling pipeline simultaneously, sharing
|
|
each round's query-response cycle across all in-flight transactions.
|
|
|
|
The key challenge is ensuring that pipelining does not compromise safety. When multiple
|
|
transactions are sampled in the same round, their confidence counters may interact
|
|
(e.g., via shared Byzantine responses). We prove that this interaction is benign: the
|
|
confidence dynamics for independent conflict sets are statistically independent, and
|
|
for transactions within the same conflict set, pipelining reduces to sequential processing
|
|
of the conflict resolution.
|
|
|
|
\paragraph{Contributions.}
|
|
\begin{enumerate}[nosep]
|
|
\item The Photon pipelined consensus protocol with confidence matrix formalization (\S\ref{sec:photon-protocol}).
|
|
\item A proof that pipelined safety degrades at most linearly in pipeline depth (\S\ref{sec:photon-safety}).
|
|
\item Pipeline stall analysis and resolution bounds (\S\ref{sec:photon-stalls}).
|
|
\item Throughput analysis showing sustained finalization at the transaction arrival rate (\S\ref{sec:photon-throughput}).
|
|
\item Empirical evaluation achieving 10,000 TPS (\S\ref{sec:photon-evaluation}).
|
|
\end{enumerate}
|
|
|
|
%% -----------------------------------------------------------------------
|
|
%% 2. SYSTEM MODEL
|
|
%% -----------------------------------------------------------------------
|