Abstract
Can we use machine learning to assist one of the most critical components of an operating system: the CPU scheduler?
In this talk, we explore how Rust and AI can be combined together to experiment with intelligent scheduling decisions in the Linux kernel.
To make this possible, we use scx_rustland_core, a Rust framework built on top of sched_ext, a recent kernel feature that allows replacing the default scheduler with a custom BPF program. This can open the door to offloading scheduling decisions to a trained AI model running in user space.
The result is a compact, AI-driven system that aims to mimic the complexity of the default kernel scheduler. We'll share the results of our experiments and discuss the practical implications of replacing traditional kernel heuristics with learned behavior.
