The Need for Speed
In High-Frequency Trading (HFT), "slow" code costs money. A difference of 1 millisecond can mean missing a profitable arbitrage opportunity. This project focuses on building a market making bot with a focus on extreme low latency.
System Architecture
We chose C++17 for its zero-overhead abstractions and manual memory control.
- Market Data Adapter: Connects to Exchange WebSockets (Binance/Coinbase).
- Order Book Manager: Maintains a local copy of the L2 Order Book.
- Strategy Engine: Calculates fair price and spread.
- Order Execution: Sends FIX/REST orders via a dedicated thread.
Optimization Techniques
Lock-Free Data Structures
Using generic std::mutex introduces context switching overhead. We used lock-free ring buffers (SPSC queues) to pass market data from the network thread to the strategy thread.
Kernel Bypass (Solarflare)
For the advanced version, we explored kernel bypass networking. By mapping the network card (NIC) memory directly to user space, we avoid the OS TCP/IP stack overhead, shaving off microseconds.
Strategy: Delta Neutral
The bot acted as a market maker, providing liquidity on both bid and ask sides. To avoid directional risk (exposure to Bitcoin price going down), we hedged our inventory on a futures contract, maintaining a Delta Neutral position.
Performance
The final system achieved an internal wire-to-order logic latency of < 40 microseconds, significantly faster than Python-based alternatives.