TUXAI: Machine learning within the Linux kernel configuration space

Abstract

Configuring a Linux kernel involves a multitude of options, numbering over 20,000. Prior to the TUXAI project, hundreds of thousands of Linux kernel configurations were compiled across different versions spanning a three-year period, resulting in a unique dataset. The primary objective of TUXAI is to predict the kernel size and the impact of each option based on various parameters using machine learning techniques. This presentation provides an overview of the work done since the beginning of the project, with a focus on technical decisions and addressing raised questions.