Benchmarking various implementations of readers-writer locks, std::sync::RwLock, parking_lot::RwLock, tokio::sync::RwLock.
cargo criterion --message-format=json > tmp/benchmark.json
We compare several different Rust implementations of readers-writer locks.
- pflock, based on "Reader-writer synchronization for shared-memory multiprocessor real-time systems"
- pairlock
- tokio
- sharedmutex
- widerwlock
- spin
- std::sync::RwLock
- crossbeam::ShardedLock
For Linux targets it is also possible to run the LLVM address sanitizer to detect memory issues in unsafe and C++ code. This requires the nightly version of the rust compiler, which can acquired using rustup toolchain install nightly. To show the symbols for the resulting stacktrace it is also convenient to install llvm-symbolizer, for example using sudo apt install llvm on Ubuntu. Afterwards, the tests can be executed with the address sanitizer enabled using cargo +nightly xtask address-sanitizer. Similarly, we
can also the thread sanitizer to detect data races using cargo +nightly xtask thread-sanitizer.
Code coverage can be automatically generated for the full workspace using a cargo task. The code coverage itself is generated by LLVM with source code instrumentation, which requires the preview tools to be installed rustup component add llvm-tools-preview, and grcov, which can be acquired using cargo install grcov. To execute the code coverage task use cargo xtask coverage. The resulting HTML code coverage report can be found in target/coverage.
Using the crate loom it is actually possible to test all possible
interleavings defined by the C11 memory standard to show that the shared mutex is data race free.
This can be used as follows:
RUSTFLAGS="--cfg loom" cargo test
If a violation is detected the following command can be used to show the exact trace and source code locations of read and write accesses.
LOOM_LOG=trace \
LOOM_LOCATION=1 \
LOOM_CHECKPOINT_INTERVAL=1 \
RUSTFLAGS="--cfg loom" \
cargo test --test <test_name> --release
- [1] BRAVO – Biased Locking for Reader-Writer Locks https://www.usenix.org/system/files/atc19-dice.pdf
- [2] NUMA-aware reader-writer locks https://dl.acm.org/doi/10.1145/2442516.2442532
- [3] brlocks https://lwn.net/Articles/378911/
- [4] Scalable reader-writer locks for parallel systems https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=222989
- [5] Scalable read-mostly synchronization using passive reader-writerlocks https://www.usenix.org/conference/atc14/technical-sessions/presentation/liu
- [6] Distributed Reader-Writer Mutex https://www.1024cores.net/home/lock-free-algorithms/reader-writer-problem/distributed-reader-writer-mutex
- [7] Distributed Cache-Line Counter Scalable RW-Lock http://concurrencyfreaks.blogspot.com/2013/09/distributed-cache-line-counter-scalable.html
- [8] Folly - https://github.com/facebook/folly