5 ms·

Note that all of the locks tested here are unfair, which is why they all show very high waiting variance. Until recently many mutex implementations aimed for fa

by gok 7y ago

Note that all of the locks tested here are unfair, which is why they all show very high waiting variance. Until recently many mutex implementations aimed for fairness, which made them much slower than spinlocks in microbenchmarks like this.

Diggsey 7y ago
Actually, the parking_lot mutex is fair: https://docs.rs/parking_lot/0.10.0/parking_lot/type.Mutex.html#fairness https://docs.rs/parking_lot/0.10.0/parking_lot/type.Mutex.ht... The high waiting variance is because the benchmark randomly decides which locks to take, meaning that the amount of contention is variable.
- sanxiyn 7y ago
  Note that fair locking is still expensive. parking_lot achieves both speed and fairness by starting with unfair locking and falling back to fair locking.
- matklad 7y ago
  Benchmark uses fixed seeds, there’s no inherent randomness.