HPC Resource Management & Security Analysis

Abstract

In multi-tenant High-Performance Computing (HPC) environments, resource isolation is critical for security. This research investigates vulnerabilities and builds system-level mitigations.

Methodology

• VRAM Auditing: Investigating NVIDIA VRAM memory leaks to mitigate cross-job data persistence risks.
• Slurm Hooks: Developing Slurm prolog and epilog hooks to automatically sanitize RAM, Disk, and GPU resources between jobs.
• Container Profiling: Profiling Apptainer container resource usage using Linux kernel tracing tools to enforce stricter boundaries.