papers/lightspeed-dex/sections/02-architecture.tex

\section{Architecture}


\subsection{Network Topology}

\begin{figure}[h]
\centering
\begin{verbatim}
                Financial Data Centers
                +----------------------+
                |                      |
    +-----------+----------------------+-----------+
    |           |                      |           |
    v           v                      v           v
+--------+ +--------+           +--------+ +--------+
|  NY4   | |  LD8   |           |  CC2   | |  SG1   |
| (NYC)  | |(London)|           |(Tokyo) | |  (SG)  |
+--------+ +--------+           +--------+ +--------+
    |           |                      |           |
    +-----------+----------------------+-----------+
                |                      |
            Fiber Optic Backbone (< 50ms RTT)
                |                      |
        +-------+----------------------+-------+
        |    Lux Consensus Validators          |
        |    (10ms GPU BLS finality, 4.7M TPS) |
        +--------------------------------------+
\end{verbatim}
\caption{Lightspeed DEX global topology}
\end{figure}

\textbf{Key locations}:
\begin{itemize}
    \item \textbf{NY4 (Equinix)}: 600+ financial firms, direct CME/NYSE access
    \item \textbf{LD8 (London)}: EMEA hub, LSE co-location
    \item \textbf{CC2 (Tokyo)}: Asia-Pacific, direct TSE access
    \item \textbf{SG1 (Singapore)}: APAC secondary, 24/7 coverage
\end{itemize}

\subsection{Order Flow}

\begin{algorithm}
\caption{Lightspeed Order Execution}
\begin{algorithmic}
\STATE \textbf{Phase 1}: Order Submission (< 1ms)
\STATE Trader submits encrypted order to nearest validator
\STATE Validator timestamps with GPS-synced clock
\STATE Order broadcast via dedicated fiber (NY4 $\to$ LD8 $\to$ CC2: 80ms)

\STATE \textbf{Phase 2}: Matching (< 1ms)
\STATE Orders collected in 10ms batches (GPU BLS finality window)
\STATE GPU-accelerated matching on sharded order books (434M orders/sec)
\STATE Price-time priority enforced

\STATE \textbf{Phase 3}: Consensus (10ms)
\STATE Matched trades bundled into 1B-gas block ($\sim$47,619 txs/block)
\STATE Quasar GPU BLS consensus (single-slot finality)
\STATE Cryptographic proof of execution order

\STATE \textbf{Phase 4}: Settlement (< 10ms)
\STATE State updates atomic via multi-sig
\STATE Cross-chain transfers initiated
\STATE Confirmation returned to trader

\STATE \textbf{Total Latency}: 22ms co-located (order $\to$ settlement)
\end{algorithmic}
\end{algorithm}

\subsection{Order Book Sharding}

To achieve 434M orders/sec, we shard by trading pair with GPU-accelerated matching:

\begin{equation}
\text{Shard}(pair) = \text{Hash}(pair) \mod N_{\text{shards}}
\end{equation}

\textbf{Example}:
\begin{itemize}
    \item Shard 0: LUX/USDC, ETH/BTC
    \item Shard 1: BTC/USDC, SOL/USDC
    \item Shard 2: AVAX/USDC, MATIC/USDC
    \item ...
    \item Shard 255: Long-tail pairs
\end{itemize}

Each GPU shard processes $\sim$1.7M orders/sec $\to$ 256 shards = 434M orders/sec.