OrderBookEngine

A high-performance, modular order book engine written in C++17. This project simulates exchange-style order matching and demonstrates practical system architecture, data structure mastery, and modern C++ practices—ideal for backend and quantitative developer interviews.

---

✨ Features

• 🧾 Order Book

  • Supports BUY and SELL limit orders
  • Maintains price-time priority via custom comparators (std::set)
  • Fast order ID lookups using std::unordered_map

• ⚙️ Matching Engine

  • Decoupled matching logic via a dedicated MatchingEngine class
  • Executes partial or full fills based on bid/ask priority
  • Returns structured Trade objects with buyer/seller IDs, price, quantity, and timestamp

• 🧪 Test-Ready Architecture

  • Structured to support Google Test and CMake integration
  • Includes a runnable main.cpp demonstrating the matching engine in action

---

📈 Benchmark: Matching Engine Performance

🔧 Environment

• C++17, STL-based std::set + unordered_map
• Single-threaded execution
• Synthetic order flow: 50% BUY @ 105, 50% SELL @ 95
• Platform: Local workstation (Intel Xeon E5-1650 v3 / Windows / MSVC)

---

⚡ Summary (v3: Batch Matching Latency)

Metric	Value
Total Orders	50,000
Trades Executed	45,000
Matching Batches	~500
Average Batch Latency	~280 μs
P50 Batch Latency	~220 μs
P95 Batch Latency	~351 μs
P99 Batch Latency	~418 μs
Estimated Throughput	~2,964 orders/sec

---

📉 Batch Latency Trend

Each call to engine.match() processes a batch of trades. We recorded per-batch latency using steady_clock::now() and plotted trends across 500+ match iterations.

Benchmarked using benchmark_orders_v3.cpp. Exported latency data is saved as CSV and visualized via Python/Matplotlib.

---

📝 Sample Output (CSV format)

Batch#, Latency(μs)
1,184
2,203
3,240
...
500,412

---

📈 Multithreaded Benchmark (Xeon E5-1650v3, 20K Orders)

Threads	Insert Time (ms)	Match Time (ms)	Total Time (ms)	Trades	Throughput
4	368	2777	3146	18000	6,356
8	309	3073	3382	18000	5,912
12	287	3100	3388	18000	5,902
16	291	3099	3391	18000	5,898

🧠 match() logic is single-threaded; further scaling requires concurrency-aware matching or lock-free queues.

---

⏱ Performance vs Thread Count

---

⚡ Throughput vs Thread Count

---

⚡ Order Latency Histogram

---

📦 Repo Structure

benchmarks/
├─ benchmark_orders.cpp // Simple throughput + match count
├─ benchmark_orders_v2.cpp // Adds per-trade latency + P95
└─ benchmark_orders_v3.cpp // Adds per-batch latency, CSV output
doc/
├─ latency_histogram.png
└─ latency_stability.png

---

✅ Next Steps (Optional)

• Add multithreaded MatchingEngine with fine-grained locks
• Add ZeroMQ or gRPC real-time gateway interface
• Plug into FIX engine and replay LOBSTER real-world L2 feeds

---

🎯 Project Goals

This engine was built as a foundational simulation for:

• Interview preparation (Google, Citadel, HRT, Jane Street, etc.)
• Showcasing backend development skills in C++
• Practicing real-time matching logic used in trading infrastructure
• Long-term expansion toward quantitative strategy simulation and HFT modeling

---

👤 Author

Qingwei Zhang
Backend Developer | Aspiring Quantitative Developer
GitHub Profile