LMAX Disruptor — High-Performance Inter-Thread Messaging for Java

Introduction

The LMAX Disruptor is a ring-buffer-based messaging framework created by LMAX Exchange to power their financial trading platform. It replaces traditional queues with a mechanical-sympathy design that minimizes lock contention, false sharing, and garbage collection pressure.

What the Disruptor Does

• Provides a lock-free ring buffer for passing events between producer and consumer threads
• Achieves single-digit microsecond latencies for inter-thread communication
• Supports multiple consumer patterns: pipeline, diamond, and broadcast topologies
• Eliminates garbage collection pauses by pre-allocating event objects in the ring
• Delivers predictable, low-jitter throughput under sustained high load

Architecture Overview

The Disruptor uses a fixed-size ring buffer backed by a pre-allocated array. Producers claim slots via an atomic sequence counter, write data, and publish. Consumers track their own sequence cursors and spin-wait or use a WaitStrategy for new events. The design avoids locks, minimizes cache-line contention through padding, and keeps allocation on the hot path to zero.

Self-Hosting & Configuration

• Add the Maven dependency to your Java project (JDK 11+ for v4.x)
• Choose a WaitStrategy: BusySpinWaitStrategy for lowest latency, SleepingWaitStrategy for lower CPU
• Set the ring buffer size to a power of two for optimal index masking
• Wire event handlers in the desired topology using handleEventsWith() and then()
• No external services or configuration files are required

Key Features

• Mechanical-sympathy design aligned with CPU cache architecture
• Pre-allocated, GC-free event passing in the hot path
• Pluggable wait strategies to trade latency for CPU usage
• Supports batching: consumers process ranges of events efficiently
• Battle-tested in production at LMAX Exchange handling millions of orders per second

Comparison with Similar Tools

• java.util.concurrent queues — ArrayBlockingQueue and LinkedBlockingQueue use locks; the Disruptor is lock-free and faster
• Chronicle Queue — persistent, memory-mapped queue for inter-process messaging; the Disruptor is in-process only but lower latency
• Aeron — UDP-based messaging for inter-machine IPC; the Disruptor focuses on intra-JVM threading
• Akka Actors — actor-based concurrency model; the Disruptor is a lower-level primitive for maximum throughput

FAQ

Q: When should I use the Disruptor over standard Java queues? A: When inter-thread latency and throughput are critical, such as trading engines, event sourcing, and log aggregation pipelines.

Q: Does it support multiple producers? A: Yes. Use ProducerType.MULTI when constructing the Disruptor for safe concurrent publishing.

Q: What happens when the ring buffer is full? A: The producer blocks (or spins) until a consumer advances, applying natural back-pressure.

Q: Is it used outside of finance? A: Yes. Log4j2 uses the Disruptor for its async loggers, and it appears in game servers, messaging systems, and data pipelines.

Sources

• https://github.com/LMAX-Exchange/disruptor
• https://lmax-exchange.github.io/disruptor/