The controversy surrounding the choice of multi-threaded executors in the Rust community is addressed in this article. Some Rust users are unhappy with this decision, believing it is a premature optimization that adds unnecessary complexity to their code. These users advocate for an alternative architecture called “thread-per-core,” claiming it is both more performant and easier to implement. However, the author argues that these two goals are in tension and that work-stealing, a feature implemented by async runtimes, is a necessary mitigation to improve tail latency and CPU utilization. The article also discusses the benefits of a share-nothing architecture and the challenges in implementing it.
https://without.boats/blog/thread-per-core/