TuxBot: Semantic-Aware Online OS Tuning with Large Language Models

arXiv cs.AI · 18 hours ago

TuxBot is a framework that uses large language models to automatically tune Linux operating system parameters on running servers by analyzing system metrics and configuration history. The system evaluated on 13 real workloads improved performance by 72.5% over default settings and 153.3% over non-LLM baselines while keeping model costs to approximately $0.20 per 30-window tuning session. The approach enables safe OS optimization with bounded model guidance through typed validation before applying kernel changes, avoiding the severe performance degradation that structure-blind parameter exploration can cause.