OrderBook Engine

Description: FIFO price-time priority order matching engine in Rust — 10M orders/sec, P99 < 200ns Topics: rust matching-engine low-latency orderbook trading fintech

• To get detail explanation of this Project I published in Medium https://medium.com/@raghavaashok2004/building-a-nanosecond-order-matching-engine-in-rust-4a7456452446

Benchmark Results

• order-book/limit-order-no-match:

• time: [95.543 ns 97.853 ns 100.03 ns]

• thrpt: [9.9968 Melem/s 10.219 Melem/s 10.466 Melem/s]

• change:

• time: [-16.775% -10.863% -4.9920%] (p = 0.00 < 0.05)

• thrpt: [+5.2543% +12.187% +20.156%]

• Performance has improved.

• order-book/limit-order-with-match

• time: [194.73 ns 195.23 ns 195.71 ns]

• thrpt: [5.1095 Melem/s 5.1223 Melem/s 5.1352 Melem/s]

• change:

• time: [-0.9761% -0.3319% +0.2483%] (p = 0.30 > 0.05)

• thrpt: [-0.2477% +0.3330% +0.9857%]

• No change in performance detected.

Architecture

• BTreeMap price levels — sorted iteration free, no per-tick sort
• SmallVec[Order; 4] — stack allocation for levels with ≤4 orders
• Zero-allocation hot path — fill buffer reused via std::mem::take
• O(1) cancel — secondary index maps OrderId → (Side, Price)
• Integer prices — stored as cents, avoids float comparison bugs

Run it

cargo run        # correctness demo
cargo bench      # benchmark with criterion
cargo flamegraph --bench throughput   # perf flamegraph

Dashboard

cargo run
# open http://localhost:3000

Live WebSocket dashboard showing throughput, P99 latency, and order book snapshot.

What's next

• FIX 4.2 protocol parser
• Lock-free SPSC input queue
• Deterministic replay engine over memory-mapped log