Zero-Allocation Limit Order Book

A limit order book and matching engine in Rust for high-frequency trading. Processes 44 million orders per second on Apple M4.

Features

• Price-time priority matching
• 64-byte cache-aligned order structs
• Pre-allocated slab memory (16 million orders, 1GB)
• Lock-free SPSC ring buffer for message passing
• Zero heap allocation on the hot path

Build

cargo build --release

Test

cargo test

Benchmark

cargo bench

Performance

Measured on Apple M4:

Metric	Result
Throughput	44.3 million orders/sec
Order struct size	64 bytes
Slab capacity	16,777,216 orders

Project Structure

src/
├── lib.rs           Module exports
├── order.rs         Order struct (64-byte aligned)
├── slab.rs          Pre-allocated memory arena
├── price_level.rs   Price level aggregation
├── orderbook.rs     Bid/ask matching logic
├── ring_buffer.rs   Lock-free SPSC queue
└── engine.rs        Matching engine

Technical Details

Order Layout

The Order struct fits exactly in one cache line:

Field	Type	Bytes
id	u64	8
price	u64	8
quantity	u64	8
remaining	u64	8
timestamp	u64	8
next	u32	4
prev	u32	4
side	u8	1
order_type	u8	1
padding	[u8; 6]	6
Total		64

Slab Allocator

Pre-allocates all memory at startup. Uses 32-bit indices instead of 64-bit pointers. Lock-free allocation via compare-and-swap.

Ring Buffer

Single-producer single-consumer queue with Release/Acquire memory fences. Cache-padded indices prevent false sharing.

License

MIT