AFP33/mpmc_event_bus

Introduction

In real-systems (logging, telemetry, message pipelines, metrics, trading systems), many threads produce events and one or more consume them. Mutex-based queues cause:

• lock contention
• priority inversion
• cache-line bouncing
• latency spikes (p99+)

This project shows how to design something better. This project is a high-performance event bus that allows multiple producers (threads or components) to publish events and multiple consumers to receive them with minimal latency and minimal synchronization overhead. It is in-process (not networked).
Think of it as a faster alternative to std::queue + std::mutex + std::condition_variable but designed for low latency, high throughput, and predictable behavior.

High-Level Architecture

• Events are written to a fixed-size ring buffer
• Producers claim slots using atomics
• Consumers read slots in order
• No global mutex on hot path

┌──────────────┐
│ Producer N   │
└──────┬───────┘
       │
       ▼
┌──────────────────────────────┐
│ Lock-Free / Low-Lock Ring    │
│ Buffer (Event Bus Core)      │
└──────┬───────────────────────┘
       │
       ▼
┌─────────────┐
│ Consumer M  │
└─────────────┘

Design Goals

1. Low latency (especially tail latency)
2. High throughput
3. Predictable performance
4. Minimal memory allocation
5. Clear ownership & lifetime rules
6. No “clever” undefined behavior

d