mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
b55171116e
Retract the fabricated '581M/597ns / 100M+/s / 434M GPU / planet-scale' headline (a time.Sleep sim + x40 fudge that measured nothing) across the corpus. Flagship lux-lightspeed-dex rewritten (decomplected d-chain engine / dexvm atomic proxy / Go wrapper / 0x9010 CLOB facade; measured GB10 104-126M & M4 Max ~250M fills/s byte-parity-gated; Knuth-D division lever; Block-STM settlement; P3Q->starkfri rollup to Z-Chain). Technical papers (precompile-engineering, gpu-crypto-stack, evmgpu-benchmark, p3q-corona, smart-order-routing, fhe-mpc-hybrid) carry the measured impl + cross-ref the flagship. >1B/s framed honestly as N-node aggregation, never single-GPU. All compile.
36 lines
2.3 KiB
TeX
36 lines
2.3 KiB
TeX
\section{Related Work}
|
|
\label{sec:related}
|
|
|
|
We position the D-Chain against the dominant on-chain trading designs along the axes that
|
|
distinguish it: where matching runs, how ordering is determined, and how finality scales.
|
|
|
|
\begin{description}
|
|
\item[Constant-product AMMs] (e.g. Uniswap~\cite{adams2020}) price every swap against a
|
|
bonding curve in a smart contract. They are simple and composable but expose order flow to
|
|
reordering and settle at block cadence with curve slippage. The D-Chain's V2/V3 contracts
|
|
are exactly such AMMs and are kept orthogonal to the CLOB facade
|
|
(Section~\ref{sec:precompile}); the matching engine itself is a CLOB, not a curve.
|
|
\item[App-specific order books] (e.g. dYdX~\cite{dydx2021}) run a CLOB with fast finality,
|
|
historically with a centralized sequencer determining order. The D-Chain also runs a CLOB,
|
|
but fixes ordering by consensus with deterministic re-derivation on every validator
|
|
(Section~\ref{sec:determinism}), so ordering is not a single sequencer's discretion.
|
|
\item[On-chain order books] (e.g. Serum~\cite{serum2020}) place the book on a general-purpose
|
|
chain, inheriting that chain's per-validator re-execution ceiling and MEV surface. The
|
|
D-Chain instead makes matching its own chain and breaks the re-execution ceiling with a
|
|
rollup that verifies one proof per $K$ trades (Section~\ref{sec:rollup}).
|
|
\item[Cross-chain derivative venues] (e.g. Injective~\cite{injective2021}) build a trading
|
|
chain with limited cross-chain reach. The D-Chain's value movement reuses the Lux primary
|
|
network's atomic shared-memory import/export (Section~\ref{sec:cross-chain}) rather than a
|
|
trusted external relayer.
|
|
\end{description}
|
|
|
|
\subsection{What is genuinely new}
|
|
|
|
The contribution is not ``a faster matching engine.'' It is the decomplection: matching and
|
|
state isolated to a chain of their own, atomic transport in a stateless proxy, an inert-by-
|
|
default EVM facade, and a rollup that separates settlement finality from matching throughput.
|
|
The performance contribution is a byte-parity-gated GPU CLOB matcher whose dominant lever is
|
|
an algorithmic one---limb-wise division---and an \emph{honest} accounting of the
|
|
$10^{9}$/s question that distinguishes per-device matching from fleet-aggregated matching and
|
|
from rollup finality.
|