Did a fun little investigation today - the smol folks reported exponential scaling in futures-concurrency.
Tracked it down, and it turned out that we could fix it by adding an extra conditional check. Outcome: at least according to this benchmark using `FutureGroup` should now perform 2-3x better than a local executor.
https://github.com/smol-rs/futures-lite/issues/93#issuecomment-2007937805
I should really write that blog post explaining why most performance-sensitive async workloads probably don’t benefit from having an executor.
That would probably be a 5/5 spice level post, as it goes against pretty much all conventional wisdom around async Rust. But I don’t think I’m wrong either.
