Скачать или смотреть Live/Historic NASDAQ ITCH Data Feed to FPGA

FPGA Trading System - Real NASDAQ ITCH 5.0 Market Data Processing

This video shows real NASDAQ historical data from December 30, 2019 being processed by an FPGA-based trading system. The demonstration covers the complete pipeline from MySQL database extraction through UDP packet transmission to hardware order book processing on a Xilinx Arty A7-100T FPGA.

SYSTEM ARCHITECTURE

The system processes 50 million NASDAQ ITCH 5.0 messages stored in MySQL. For this demo, I'm using a test dataset of 80,000 messages covering eight symbols: AAPL, TSLA, SPY, QQQ, GOOGL, MSFT, AMZN, and NVDA. The Python script maintains a sustained rate of over 600 messages per second, sending UDP packets to the FPGA which processes nine different ITCH message types.

PERFORMANCE CHARACTERISTICS

The Python script processes messages at 400+ per second using client-side cursor optimization. UDP packets are generated in binary format with big-endian encoding. On the FPGA side, the MII physical layer feeds into the UDP/IP parser, then to the ITCH decoder. Total hardware latency from wire to order book is under 5 microseconds. Individual order processing takes 120-170 nanoseconds, and Best Bid/Offer updates complete in 2.6 microseconds after scanning 256 price levels.

WHAT THE VIDEO SHOWS

The console displays real-time message statistics including Add Orders, Trades, and Executions. You'll see symbol distribution across the eight equities, UDP packet transmission rates, and FPGA order book state updates. The output shows message type distribution, price levels and share quantities, complete order lifecycle from Add through Execute, Cancel, and Delete, plus performance metrics.

TECHNICAL IMPLEMENTATION

Database is MySQL 8.0 with optimized bulk queries. The Python script uses PyMySQL and socket programming. Network communication is UDP/IP broadcast to 192.168.0.93 port 5000. FPGA is a Xilinx Arty A7-100T using VHDL for the UDP/IP stack, ITCH parser, and order book implementation. All 80,000 messages were processed with 100% accuracy during validation.

TRADING SYSTEM CAPABILITIES

The system implements eight parallel BRAM-based order books with real-time Best Bid/Offer tracking. Symbol filtering routes messages to the appropriate book. Hardware timestamping provides deterministic processing using fixed-cycle state machines.

This demonstrates a complete market data pipeline from database extraction through network transmission to FPGA hardware acceleration, which is the core infrastructure for low-latency trading systems.

GitHub: https://github.com/adilsondias-engine...
Project files: database.md, send_itch_packets.py, Projects 6-8