moodycamel::ConcurrentQueue

A widely-deployed C++ MPMC queue (18k+ GitHub stars) by Cameron Desrochers. Its performance comes from per-producer implicit sub-queues that eliminate inter-producer contention: each producer enqueues to its own sub-queue, and consumers round-robin across sub-queues. This trades strict cross-producer FIFO ordering — which is rarely actually needed — for ~1.7× throughput over Boost.Lockfree and competitive performance across the C++ MPMC landscape.

Design

When a thread first enqueues, it gets allocated an implicit producer slot. That producer's sub-queue is essentially SPSC — only that thread writes to it — so per-producer push is fast and uncontended. Consumers iterate sub-queues looking for a non-empty one and pop. The result:

• ~114M ops/s single-threaded; scales well with producers
• Sub-100 ns push and pop in the common case
• No strict FIFO across producers (a later push by producer A may be consumed before an earlier push by producer B)

For most workloads — task queues, work distribution, logging — relaxed FIFO is fine. For workloads where strict ordering across producers is required (e.g., serializing requests for a state machine), prefer LCRQ, SCQ, or atomic_queue.

Comparison

When to choose it

• General-purpose C++ MPMC where strict cross-producer FIFO is not required
• Many producers known to outnumber consumers (the implicit-producer model amortizes well)
• Pre-existing C++ codebase looking for a drop-in replacement for std::queue + mutex

For lower latency or strict FIFO, atomic-queue. For x86-only max throughput, the C++ port of lcrq.

See also

• mpmc-queue — broader landscape
• atomic-queue — strict-FIFO C++ alternative
• lcrq — the throughput ceiling
• concurrent-queues