Building a Low-Latency Trading System: Design and Implementation

This book walks through the complete architecture of Madrigal, a collection of fault-tolerant, low-latency trading system components built on the Presto infrastructure layer — covering the design decisions, trade-offs, and failure modes that only show up under real market conditions, with data structures and algorithms explored in the context of a real trading system throughout. Both Madrigal and Presto are open source and available on GitHub (https://github.com/A-Cappella-dot-Net/opus-1/), so you can follow the thought process behind every architectural decision alongside the actual working code. You'll learn how to: Implement inter-process communication primitives over TCP sockets Design distributed fault-tolerance mechanisms, including leader election algorithms and process-instance coordination and monitoring Build a pub/snap-and-subscribe API — the backbone of inter-process communication — with flexible merge strategies, snapshot-providing services, and self-throttling message conflation Implement a serializer component that guarantees serialized events arrive in the same order across all interested components Measure and analyze performance rigorously, eliminating sources of garbage to keep tail latencies predictable Build a generic exchange matching engine covering both continuous and auction phases Design an authentication and access-control architecture for trading infrastructure Implement advanced order management tactics that serve as building blocks for higher-level algorithms: queue-preserving amends, simulated iceberg and IOC orders, and sniper/hidden orders Written for Java engineers, quants, and systems architects building or maintaining trading platforms, this book pairs concrete implementation detail with the reasoning behind it — using a single open-source running system as its example throughout, so every concept connects back to code you can read, run, and extend yourself.