Designing a Stock Trading Platform

Stock trading platforms require microsecond latency, strict ordering, and fault tolerance.

Core Components

Order Entry → Risk Checks → Matching Engine → Trade Execution
                                 ↓
                          Market Data Feed

Order Types

• Market Order: Execute immediately at best price
• Limit Order: Execute only at specified price or better
• Stop Order: Trigger when price crosses threshold

Order Book

Bids (buy) and asks (sell) sorted by price:

Bids (Buy):              Asks (Sell):
Price   | Qty          Price   | Qty
$150.00 | 100          $150.10 | 200
$149.95 | 250          $150.15 | 150
$149.90 | 500          $150.20 | 300

Spread = Ask - Bid = $0.10

Matching Engine

Matches buy and sell orders:

New Buy Order: $150.10, 100 shares
Best Ask: $150.10, 200 shares
→ MATCH: 100 shares at $150.10
→ Ask quantity reduced to 100

Algorithm: Price-time priority (FIFO at same price).

Latency Requirements

HFT firms target <1 microsecond
Retail platforms target <1 millisecond

Techniques:

• In-memory order book (no DB in hot path)
• FPGA for ultra-low latency
• Kernel bypass networking (DPDK)
• Co-location at exchange data centers

Sequence Numbers

All orders tagged with monotonically increasing sequence number — ensures total order.

Market Data Feed

Broadcast order book changes:

• Level 1: Best bid/ask
• Level 2: Full depth
• Trades feed

Using multicast UDP for low-latency broadcast.

Conclusion

Matching engines are the most performance-critical systems. In-memory processing, sequential guarantees, and kernel bypass networking are essential for microsecond-level performance.