papaya

A lock-free concurrent HashMap for Rust with novel memory reclamation. The best choice for read-heavy caches with predictable latency — reads never block, and it is safe for async usage with no deadlock risk (unlike dashmap's RwLock-based design).

When to use which concurrent map

• papaya — lock-free reads, predictable latency, async-safe. Best for read-heavy caches.
• dashmap — sharded RwLocks, best write throughput, familiar HashMap API. 173M+ downloads. Best for balanced read/write workloads.
• scc — aggressive bucket-level locking for extreme write contention.

See rust-concurrent-data-structures for the full landscape.

Where Rust concurrent maps sit globally

Even Rust's best concurrent maps trail the C++ state-of-the-art at very high thread counts. CMU's parlayhash reaches 1,130 Mops at 128 threads — roughly 7× Meta's folly::ConcurrentHashMap and 39× libcuckoo. Papaya, dashmap, and scc do not have published benchmarks at that scale, but they are not architected around epoch-based reclamation in the same way ParlayHash is. For workloads with extreme thread counts where you can use C++, ParlayHash is the headline.

The other caveat: concurrent maps only earn their overhead above ~4–8 threads. Below that, a single-threaded hashbrown behind a Mutex often wins. See fastest-hash-map-2025 for the broader picture.