rw_lease
Fast Reader-Writer lock with reader draining support. Based on a single (parameterisable) atomic usize.
Notes:
- Steals the high bit for the writer lock.
- Designed for low contention, mostly-read workloads.
- Write locking requires waiting for readers to drop their locks.
Benchmarks
There are benchmarks, which you can and should run. Here are some numbers from my 2015 macbook pro on an AtomicUsize (the default):
| Benchmark | Mutex | RwLock | RWLease |
|---|---|---|---|
| Create | 110 | 107 | 1 |
| Uncontended Reads | 330731 (1.4) | 417664 (1.77) | 235656 (1) |
| Contended Reads | 1140321 (1) | 2367186 (2.08) | 1488557 (1.31) |
Notes: measurements in nanoseconds, parens = normalised to shortest run.
We haven't spent terribly long optimising the code, there may be some wins left to gain.
Copyright and License
Copyright (c) 2020 James Laver, rw_lease contributors.
This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.