I'm curious how folks in Rust implement striped-locking?
Striped locking is where you chunk sections of data behind the same lock to improve scalability. So, 1 lock might protect 1,000 items in an array, for example.
Striped locking is where you chunk sections of data behind the same lock to improve scalability. So, 1 lock might protect 1,000 items in an array, for example.
struct Striped<T, const StripeSize: usize>(Vec<Mutex<[T; StripeSize]>>, Mutex<tinyvec::ArrayVec<[T; StripeSize]>>);
Rust makes writing very high performance multi-threaded code incredibly hard because most of the tricks to get that last bit of performance out are basically asking to generated races and are near impossible to express in a way that is provably safe. You basically thinking of your data-layout from day one if you need that kind of performance your just going to have to deal with a bunch of unsafe in your codebase.
[Mutex<[T; 1000]>; 10]