User API & C++ Implementation of a Multi Producer, Multi Consumer, Lock Free, Atomic Queue

on YouTube ↗01h 03m

" summary

Lock-free MPMC queue implemented in C++17 std::atomic, uses a circular array of fixed-size, cacheline-aligned entries, no heap allocation at runtime, and a 16-byte atomic compare‑and‑swap trick (cmpxchg16b / 128‑bit CAS) to atomically flip entry state and data. Benchmarks report up to 703M msgs/s (single‑producer/single‑consumer on a laptop) and sub‑microsecond round‑trip latency; tested on Intel/AMD, supports unique_ptr wrappers for larger payloads, and is available on GitHub.

AI-written summary · about

" chapters (12)

1. 0:00Lock-free multi-producer queue intro
2. 1:29Low-latency queue requirements
3. 7:22Fat tails performance impact
4. 17:51Hardware atomics for queues
5. 22:53Atomics not always lock-free
6. 25:22Lock-free queue design
7. 31:34Queue API and examples
8. 42:28Push and pop operations
9. 46:31Testing and validation
10. 49:23Performance benchmarks
11. 51:34Future work and optimizations
12. 54:32Summary and questions

" related talks

• 01h 02mConcurrency PatternsRainer Grimm · CppCon 2021
• 20mMulti-Queue Fair QueuingMohammad Hedayati, Kai Shen, Michael L. Scott, Mike Marty · USENIX ATC '19
• 01h 02mBack to Basics: ConcurrencyMike Shah · CppCon 2021
• 01h 01mReal-time Programming with the C++ Standard LibraryTimur Doumler · CppCon 2021
• 39mUsing Race Conditions in Correct Concurrent SoftwareDevon O'Dell · Strange Loop 2015
• 01h 10mBasic usage of PMRs for better performanceMarek Krajewski · Meeting C++ 2022
• 01h 00mMeasuring Qt PerformanceJoão Abecasis · Qt Developer Days 2010 (Munich)
• 01h 01mRefresher on Containers, Algorithms and Performance in C++Vladimir Vishnevskii · CppCon 2022