Magma — Ultra-Low Latency Trading Engine

Magma is a high-performance, lock-free financial matching engine written in Core Java.
It achieves sub-microsecond latency (130ns median) on commodity hardware by leveraging the LMAX Disruptor pattern, a single-writer architecture, and GC-free memory management.

This project demonstrates low-latency systems engineering, high-performance concurrency, and deterministic execution techniques used in real-world trading systems.

---

Performance Benchmarks

Environment: AMD Ryzen 5 7530U (Single Thread, CPU Affinity Pinned)

• Throughput: ~1.2 million orders / second
• Median Latency (p50): 130 nanoseconds
• Tail Latency (p99): 3.5 microseconds

---

Architecture

Magma is built around a Single-Writer / Single-Reader architecture to eliminate lock contention and minimize memory barriers.

Core Components

1. Gateway

   • TCP server using java.nio (Selector-based)
   • Handles concurrent client connections with non-blocking I/O

2. Ring Buffer

   • Custom power-of-two circular buffer
   • O(1) slot claiming
   • Cache-friendly layout inspired by the LMAX Disruptor

3. Matching Engine

   • Price-Time Priority (FIFO) limit order book
   • TreeMap for price levels
   • LinkedList for strict time ordering

4. Object Pool

   • Pre-allocated pool for 1M+ orders
   • Zero allocations on the hot path
   • Eliminates GC pauses

---

Tech Stack

• Language: Java 21 (Core Java)
• Concurrency: Unsafe, AtomicLong, volatile semantics
• Networking: java.nio (Non-blocking I/O)
• Data Structures: Custom RingBuffer, TreeMap, LinkedList

---

How to Run

1. Start the Server

Run Main.java to start the matching engine and TCP gateway.

# Example output
Gateway Server listening on port 9090

---

2. Connect via Telnet

telnet localhost 9090

Example session:

> BUY 100 10
> SELL 100 10
< OK: SELL 10 @ 100.0

---

3. Run Benchmarks

Run BenchmarkMain.java to:

• Flood the engine with 1 million orders
• Generate throughput and latency statistics
• Produce a latency histogram (p50 / p99)

---

Project Structure

src/
├── gateway/        # NIO TCP server and protocol parsing
├── engine/         # Matching engine core
├── orderbook/      # Limit order book implementation
├── ringbuffer/     # Lock-free ring buffer
├── pool/           # Object pooling & memory reuse
├── benchmark/      # Load & latency benchmarks
└── Main.java       # Entry point

---

Design Goals

• Zero locks on the hot path
• Single-writer determinism
• Cache-friendly memory layout
• Predictable latency under load
• No GC activity during trading

---

Git Setup & Push Instructions

git init
git add .
git commit -m "Initial commit: ultra-low latency matching engine"
git branch -M main
git remote add origin https://github.com/BAT4K/Magma-Trading-Engine.git
git push -u origin main

---

Author

Hans James

---

License

This project is released under the MIT License.