SSC24/25

STELLAR: Storage Tuning Engine Leveraging LLM Autonomous Reasoning for High Performance Parallel File Systems

Inspired and motivated by the availablity and accessibility of real cluster hardware Chris Egersdoerfer and his collaborators Robert Ross, Philip Carns, and Shane Snyder from Argonne National Lab advanced their efforts to integrate AI agents into HPC storage systems. STELLAR introduces an agent-based tool that autonomously tunes parallel file systems configurations. It tailors settings iteratively and insightfully for diverse workloads. The paper has been accepted for the SC25 and the tool will be released in the near future.