mirror of
https://github.com/lux-apps/papers.git
synced 2026-11-03 05:07:46 +00:00
zeekay
f4ee7a0525
Bylines, paper authorship, correspondence addresses and the founder blurb under each post named an individual. They name the organisation that published the work instead: the brand on posts and pages, the research byline on papers, and research@ in place of a personal address. Both are what these repositories already used elsewhere — nothing new was coined. Co-authors are untouched. Where a paper named someone else alongside, they stay named; only the one byline changed. Bibliography entries, running heads, PDF metadata and affiliations move with it, so a citation still resolves to whoever the cited paper names. Sample and fixture data uses the placeholder name these tests already use elsewhere, so no assertion changed meaning.
110 lines
4.2 KiB
TeX
110 lines
4.2 KiB
TeX
\documentclass[11pt,a4paper]{article}
|
|
\usepackage{../shared/luxcover}
|
|
\usepackage[utf8]{inputenc}
|
|
\usepackage[T1]{fontenc}
|
|
\usepackage{amsmath,amssymb,amsfonts,amsthm}
|
|
\usepackage{graphicx}
|
|
\usepackage{hyperref}
|
|
\usepackage{booktabs}
|
|
\usepackage{algorithm}
|
|
\usepackage{algpseudocode}
|
|
\usepackage{enumitem}
|
|
\usepackage{xcolor}
|
|
\usepackage{listings}
|
|
\usepackage{array}
|
|
|
|
\hypersetup{colorlinks=true,linkcolor=black,citecolor=blue,urlcolor=blue}
|
|
|
|
\newtheorem{theorem}{Theorem}[section]
|
|
\newtheorem{lemma}[theorem]{Lemma}
|
|
\newtheorem{corollary}[theorem]{Corollary}
|
|
\newtheorem{proposition}[theorem]{Proposition}
|
|
\newtheorem{definition}{Definition}[section]
|
|
\newtheorem{invariant}{Invariant}[section]
|
|
\newtheorem{remark}{Remark}[section]
|
|
|
|
\lstset{basicstyle=\ttfamily\small,breaklines=true,columns=fullflexible,
|
|
keywordstyle=\bfseries,commentstyle=\color{gray!70!black}\itshape}
|
|
|
|
\title{LP-010: QuasarSTM\\[4pt]
|
|
\large Block-STM 3.0 --- A GPU-Native Lane-Aware\\
|
|
Ordered MVCC Fabric for Quasar-Certified Rounds}
|
|
|
|
\author{
|
|
Lux Research\thanks{Corresponding author: [email protected]}\\
|
|
\textit{Lux Industries, Inc.}\\
|
|
\texttt{[email protected]}
|
|
}
|
|
|
|
\date{Spec freeze: December~15, 2025\quad|\quad Activation: December~25, 2025}
|
|
|
|
\begin{document}
|
|
|
|
\luxcoverpage
|
|
|
|
\begin{abstract}
|
|
We present \emph{QuasarSTM}, the third generation of Block-STM, designed
|
|
as the ordered multi-version concurrency control (MVCC) fabric beneath
|
|
Lux's Nova (linear) and Nebula (DAG) execution modes. Block-STM~1.0
|
|
(Aptos~2022) introduced ordered speculative execution on CPU.
|
|
Block-STM~2.0 was the GPU wave-drain port, treating the GPU as a
|
|
faster CPU. QuasarSTM~3.0 is a re-architecture: STM is no longer a
|
|
retroactive conflict detector; it is a scheduler-aware fabric. Its
|
|
primary abstraction is the \emph{lane}---a deterministic hash of
|
|
\textsf{(domain, contract, account, asset, market, nonce-lane,
|
|
storage-prefix)}---which is the smallest independently schedulable
|
|
state domain. Lanes drive Prism refraction (so STM rarely sees the
|
|
same hot conflict twice), three-tier validation (lane-clock, key-MVCC,
|
|
semantic commutativity), deterministic contention management, and
|
|
multi-GPU sharding.
|
|
|
|
We give formal definitions of lanes, ordered MVCC visibility for
|
|
both Nova and Nebula, three-tier validation, TicToc-style per-key
|
|
timestamps that reduce false aborts without reordering consensus,
|
|
checkpoint-based incremental repair, and the commit horizon. We
|
|
prove four invariants---determinism, ordered serializability, repair
|
|
monotonicity, and horizon safety---and report production-hardware
|
|
measurements: a 1024-transaction end-to-end stress completes in
|
|
${\sim}150\,\text{ms}$ on Apple~M1~Max in 8~wave ticks, and the canonical
|
|
hot-key repair workload (16~same-key contending transactions)
|
|
yields exactly 120~conflicts, 120~repairs, and 16~commits---textbook
|
|
Block-STM running entirely on GPU. The headline target is
|
|
$\mathit{repair\_amplification} < 1.01$ on normal regulated-DEX
|
|
workloads, with controlled degradation rather than repair explosion
|
|
under heavy contention.
|
|
\end{abstract}
|
|
|
|
\tableofcontents
|
|
\newpage
|
|
|
|
%% ============================================================================
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-background}
|
|
\input{sections/03-lanes}
|
|
\input{sections/04-three-tier-validation}
|
|
\input{sections/05-visibility-rule}
|
|
\input{sections/06-tictoc-timestamps}
|
|
\input{sections/07-prism-refraction}
|
|
\input{sections/08-checkpoint-repair}
|
|
\input{sections/09-contention-manager}
|
|
\input{sections/10-commit-horizon}
|
|
\input{sections/11-gc-multi-gpu}
|
|
\input{sections/12-correctness}
|
|
\input{sections/13-performance}
|
|
\input{sections/14-related-work}
|
|
\input{sections/99-conclusion}
|
|
|
|
\bibliographystyle{plain}
|
|
\bibliography{lp-010-quasar-stm-3}
|
|
|
|
\appendix
|
|
\section*{Reproducibility}
|
|
\textbf{Source code.} \texttt{github.com/luxfi/consensus} (commit \texttt{TBD}); GPU kernels at \texttt{github.com/luxfi/gpu}; STM runtime at \texttt{github.com/luxfi/stm}.\\
|
|
\textbf{Build.} \texttt{go build ./...}; GPU paths require CUDA \(\geq\) 12 or Metal toolchain on macOS.\\
|
|
\textbf{Hardware tested.} TBD (multi-GPU x86\_64 Linux; Apple Silicon for the Metal back-end).\\
|
|
\textbf{Standards referenced.} FIPS 204 (ML-DSA) for the PQ lane; project-internal LP-010, LP-020, LP-070.\\
|
|
\textbf{Contact.} [email protected].
|
|
|
|
\end{document}
|