Single-Threaded · C++17 · Linux epoll

Limit Order Book
Matching Engine

A production-grade C++ matching engine built in four phases. From a correct in-memory core to a hardware-pinned, nanosecond-latency networked exchange.

Simulation Note: The interactive widgets below are a pure JavaScript implementation of the engine logic for visualization. To run the actual engine with true nanosecond latency and zero-allocation guarantees, clone and compile the C++ codebase.

<3μsMedian Latency

1MPre-allocated Slots

14BWire Protocol

0Heap Allocs on Hot Path

Try Simulation Read the Docs Project Architecture

Interactive Demo

This is a faithful JavaScript implementation of the C++ matching engine running in your browser. Price-time priority, partial fills, cancellations — all exact.

Place Order

Price ($)

Quantity

Cancel Specific Order

Auto Simulation

Speed (ms) 600ms

Order Book Ladder

LIVE

BID QTY PRICE ASK QTY

Spread: —

Execution Tape

Memory Pool State

The engine pre-allocates one million OrderNode slots at startup. Every order allocation and deallocation is an O(1) array index operation — no OS heap calls on the hot path.

Pool Statistics

Used Slots 0%

0Active Orders

1,000,000Free Slots

0Next Free Idx

Free List Head Pointer

Occupied Slots

The physical indices shown are the actual array positions in the one-million-element arena that currently hold live orders. When an order is cancelled the slot returns to the free-list head and will be handed out first on the next allocation.

Node Data — Raw OrderNode Structs

Every resting order is stored as an OrderNode in the pool arena. The prev_idx and next_idx fields form the doubly linked list that encodes time priority at each price level.

Active Nodes (first 10)