Design of a Unified Monitoring Tool for Detecting Anomalies in High Performance Computing Systems

With increased demand in various cyberinfrastructure, researchers are leveraging High Performance Computing (HPC) systems to simulate complex environments for system and cybersecurity experimentation. This paper presents ongoing work towards designing a novel unified monitoring framework that collects both system-level and network-level metrics from containerized networking system orchestrated in a public cloud environment, to support anomaly detection in HPC systems. The framework captures system-level metrics using Lightweight Distributed Metric Service (LDMS), further hardware performance data through Performance Application Programming Interface (PAPI), and network flow features using tcpdump. This paper mainly focuses on the design of the framework and a case study associated with the effectiveness of some of the collected metrics.