Performance Comparison of Mutex, RWLock and Atomic types in Rust
7 likes11,985 views
This document compares the performance of Mutex, RWLock, and atomic types for concurrent access in Rust. It presents benchmark results from a Rust program that tests updating a shared variable with these different synchronization methods under varying read/write loads on Linux and macOS. The results show that RWLock performance degrades significantly with more writes, Mutex performance is consistent even with writes, and atomic types are always the fastest for a counter.
Performance Comparison of Mutex, RWLock and Atomic types in Rust
- 1. Performance Comparison of Mutex, RWLock and Atomic types in Rust Mitsunori Komatsu
- 2. About me Mitsunori Komatsu, Software engineer Started learning Rust recently (7 months ago) Contributed a small feature to Rust (https:// github.com/rust-lang/rust/pull/33976) https://github.com/komamitsu/td-client-rust https://github.com/komamitsu/treasuredata_fdw
- 3. It’s hard to update a variable serializably in multi-threads…
- 4. Mutable Types in Arc (Arc uses atomic reference count) Mutex Mutual Exclusion Lock Can contain Send RWLock Read: Shared Lock Write: Exclusive Lock Can contain Send + Sync Atomic types Atomic counter - We can update these types in multi-threads
- 9. Benchmark program (Atomic types #1) https://github.com/komamitsu/rust-lock-bench
- 10. Benchmark program (Atomic types #2) https://github.com/komamitsu/rust-lock-bench
- 11. Let’s Benchmark! Rust 1.15.1 Linux 4.4 (c3.2xlarge) macOS 10.11
- 12. Benchmark result (OS: Linux, Build: Debug, Loop: 50000, Total Threads: 20, # of trial: 20)
- 13. Benchmark result (OS: Linux, Build: Release, Loop: 50000, Total Threads: 20, # of trial: 20)
- 14. Benchmark result (OS: macOS, Build: Debug, Loop: 10000, Total Threads: 20, # of trial: 20)
- 15. Benchmark result Benchmark result (OS: macOS, Build: Release, Loop: 10000, Total Threads: 20, # of trial: 20)
- 16. Summary The performance of RWLock expectedly degrades when the number of writes increases The performance of Mutex doesn't significantly change even under write heavy operations. Also, it’s really optimized on Linux AtomicUsize is always really fast. Use it if you want just a counter