SAMPA: Security Analysis and Monitoring to Prevent Abuse of HPC Environments

In general, HPC systems should not be used to carry out cybercrimes or execute improper jobs- Existing fingerprinting-based solutions can be used to understand what are running on HPC systems, but they typically are reactive, require system-wide data collection and processing, and do not consider adversarial scenarios- Hence, it is essential to build offline code analysis tools that can provide first-line defend against (adversarial) illegal and improper HPC programs submitted by users accidentally or intentionally, and take preemptive action to prevent the abuse of HPC systems- To address this critical need, a semantic-aware code analysis toolset is proposed to understand the behaviors of different programs and detect illicit or improper programs- The overall project approach is compatible with existing HPC compiler infrastructure with little integration effort, works with many programming languages used in HPC environments such as C/C++ and Fortran, and performs static (offline) analysis without runtime performance overhead-