A small in-memory price-time-priority limit order book for Rust, intended as a matching core for exchanges with straightforward rules.
submit: matches an order and returns its maker fillscancel: removes an open order by identifierreduce: lowers an order's open quantity without changing prioritybest_bid,best_ask, and iterators inspect resting orders- Application-owned order, identifier, quantity, price, and side representation
- Immediate-or-cancel: call
submit, thencancel - Add-liquidity-only: compare the submitted price with
best_askorbest_bid - Stop orders: hold the order outside the book until its trigger fires
- Each open order has a unique identifier. An identifier may be reused after its previous order leaves the book.
- Quantities represent non-negative magnitudes. Zero is valid, including for fills and reductions. The caller validates this requirement.
- Subtracting a smaller quantity from a larger quantity produces a valid non-negative remainder.
- Prices have a lawful, stable total ordering. They do not need to be numeric or positive.
- An order's identifier, price, and side remain stable while it rests.
set_quantitychanges only its quantity.
VecBookis the reference implementation. It stores each side in a sorted vector.LevelBookmaps prices through balanced trees, stores orders and levels in generational vector arenas, and indexes identifiers for direct lookup, cancellation, and reduction.FlatLevelBookuses the same arena-backed FIFO levels and identifier index, but keeps prices in sorted vectors. It targets books with relatively few active price levels.
On one thread of an AMD Ryzen 9 9950X, FlatLevelBook processes roughly 50 million
book actions per second across the synthetic workloads. Each workload replays 100,000
submissions, cancellations, and reductions; one submission may execute several resting orders.
| Workload | VecBook |
LevelBook |
FlatLevelBook |
|---|---|---|---|
| Mixed | 0.6M/s | 37.3M/s | 49.0M/s |
| Cancel-heavy | 0.5M/s | 45.0M/s | 57.3M/s |
| Sweep-heavy | 2.0M/s | 36.4M/s | 42.9M/s |
| Empirical | 1.2M/s | 44.4M/s | 52.3M/s |
These are Criterion mean throughputs measured with Rust 1.95.0 and
-C target-cpu=native. Results depend on hardware, compiler, book shape, and action mix. See the
benchmark notes for workload definitions and commands.
Choose another matching engine when order behavior must run atomically inside the matching loop and cannot be expressed through limit submissions, cancellations, and reductions. This includes:
- pro-rata, size-priority, auction, or other non-price-time matching
- native iceberg refresh, pegged, discretionary, stop, or conditional-order state machines
- atomic multi-leg, spread, or cross-book matching
Rich order structs are supported because the book carries their fields unchanged. Matching behavior remains limited to price-time-priority limit orders.
use lobster::{Fill, OrderBook, SimpleOrder, VecBook};
let mut book = VecBook::default();
let _ = book.submit(SimpleOrder::sell(0, 2, 5));
let maker = SimpleOrder::sell(1, 3, 6);
let _ = book.submit(maker);
let _ = book.submit(SimpleOrder::sell(2, 4, 7));
book.cancel(&0);
let fills = book.submit(SimpleOrder::buy(3, 6, 6));
assert_eq!(fills, [Fill::Full(maker)]);