Congee

Rust port of ART with Optimistic Lock Coupling. Hits 150 Mops/sec on 32 cores with 8-byte integer keys, putting it in the same tier as the C++ reference implementation and far ahead of dashmap / scc / papaya for ordered concurrent workloads (those are unordered hash maps and live in a different design space — see fastest-hash-map-2025).

Why it exists

The Rust ecosystem has excellent concurrent unordered maps but until Congee had no production-quality concurrent ordered map matching the C++ state of the art. BTreeMap is single-threaded; wrapping it in RwLock collapses under contention. Congee fills the same niche parlayhash fills on the unordered side: the implementation that closes the gap to the global frontier.

When to reach for it

• Concurrent point operations on integer or fixed-width keys.
• Workloads that need ordered traversal (otherwise prefer papaya or scc for read-heavy / write-heavy hash-map paths).
• High thread counts where lock-coupled BTreeMap wrappers melt down.

For range-scan-heavy concurrent workloads the global state of the art is bp-tree (no current Rust port). For string-key high-contention workloads it's masstree (also unported). Congee is the closest the Rust ecosystem currently sits to the concurrent-ordered-map frontier — see rust-concurrent-data-structures and fastest-ordered-maps.