English version first. 中文版本請見下半部 (跳到中文版).
Background
In late 2024, the ZIO project opened issue #9356 with a $2,500 bounty: re-implement ZIO’s task scheduler using the design ideas from Nio, a 150-line experimental Rust runtime that claimed to outperform Tokio’s work-stealing scheduler on several benchmarks by replacing the random-stealing strategy with a “least-loaded worker” dispatch.
I picked up this bounty in February 2026. My background is Node.js + C/C++ system programming, not JVM. After three months of implementation, benchmarking, and reading other contributors’ attempts, I concluded that the bounty as literally specified is structurally hard to deliver on the JVM, and the difficulty is not algorithmic — it is a memory-model and architectural mismatch. This post is a retrospective for myself, and a reference for anyone considering the same problem.