The Performance Cost of RwLock in Our Read-Heavy Workload
Summary
The article analyzes the performance costs of using RwLock in a read-heavy Rust workload and compares it to a lock-free approach using Crossbeam epoch-based pointers. It presents a data model and benchmark results showing that lock-free achieves far higher read throughput and lower latency, but acknowledges that RwLock can still be effective when used around larger structures. The piece emphasizes that where and how synchronization occurs drives performance, and it reviews alternatives like ArcSwap and left-right.