flume VS async-wormhole

The repository seems abandoned; or maybe complete?

At work we use flume, which is another capable multi-producer, multi-consumer async-capable channel [1]. It's great for shuffling data between threads, as well as between async tasks, and between threads and async tasks. Basically any time you want to pieces of code to exchange data or signals without pesky shared state.

1: https://github.com/zesterer/flume

I also like flume, it has impressive performance (although not the best). More importantly, it's written only with safe rust. https://github.com/zesterer/flume

The most commonly suggested replacement for mspc is crossbeam-channel; flume is also relatively popular.

It's not officially deprecated, but the alternatives on crates.io are considered better. flume and crossbeam-channel feature less unsafe code and offer better performance. Benchmarks.

it's slow (checkout flume's benchmarks for example)

I would have done the same. I think, and I might be wrong, but the only other alternative, besides anything unsafe, would be to pass mutex back, but I am not sure this would be faster. Btw, I have not done testing, but you might want to look at Flume for your mpsc channels: https://github.com/zesterer/flume Flume, seems to be very fast mpsc implementation. I am planning to evaluate it for logging system.

For concurrency/parallelism, you launch at most 2 * CPU Cores, PIN them and use a fast broker to spread the task (like a ring buffer or an MPSC). But you keep linear scan, tight loops, SIMD friendly data, on each. You are not switching context that much, and instead, bet you will process the batch fast. (CPUs are fast today!)

https://crates.io/crates/flume and https://crates.io/crates/crossbeam-channel provide MPMC channels.

Taking onto such a big project can be scary and overwhelming, so I like to "cheat" a bit. Instead of developing a M:N scheduler I picked an already mature and proven one from the Rust ecosystem: tokio. Then I just needed to develop a virtual stacks solution that works well with the scheduler. Instead of inventing my own byte-code I just picked WebAssembly, it's just a small abstraction above machine code and has mature JIT compiler libraries that generate code close to native speed. Then again, I just needed to figure out how to do reduction counting and insert preemption points into WebAssembly code during loading.

Previously, we used our own implementation of virtual stacks and stack switching. Both (Wasmer & Wasmtime) Wasm runtimes we used internally required a tight integration with it, but neither library exposed the primitives to integrate well with it. So we needed to maintain forks of both runtimes with some patches to expose internal data structures. Just keeping up to date with new releases was taking way too much of my time. Stack switching is also a delicate task with a lot of hand written assembly involved and we would run into segfaults from time to time. Luckily Wasmtime shipped "native" async support that works similar to our implementation so we could switch to it. Re-writing was a pain, but I'm happy how everything has turned out and think that it was worth it.

Under the hood, Lunatic wraps "processes" inside of Rust async tasks with https://github.com/bkolobara/async-wormhole and can use any async executor to run them. We are currently using Smol's multithreaded executor, so it scales quite nicely across cores.