Two NSF awards granted to NCSA scientist Phuong Cao and ECE Professor Ravishankar Iyer

Phuong Cao, a research scientist at the National Center for Supercomputing Applications (NCSA) and Ravishankar K. Iyer in the Grainger College of Engineering at the University of Illinois Urbana-Champaign recently received two National Science Foundation (NSF) grants. The first award will advance efforts in curating security operational data for open-science research, which is the main driver for the second award on identifying misuse of HPC resources that will influence research security policies. Both awards reflect NSF’s primary focus on safeguarding the integrity and security of science while also keeping fundamental research open and collaborative.

Iyer, the George and Ann Fisher Distinguished Professor of Engineering in electrical & computer engineering and the Coordinated Science Laboratory, is working with graduate student Archit Patke on one resiliency project and cybersecurity researcher Phuong Cao at NCSA on the other cybersecurity projects. While the two projects share the same high-level goal, they are approaching it from different angles.

Cao’s security research provides a decade-long study of cyberattacks targeting NCSA, showing a 360-view of threats targeting open-science research in U.S. cyberinfrastructure. He gained a deep understanding of how historical backdoors evolved into cutting edge threats, leveraging AI- and possibly Quantum-driven techniques to exploit vast HPC/AI resources and leak embargoed scientific data. Grounded by security measurements provided by Zeek, he has been developing attack preemption models, not only capturing evolution of known attacks on High Performance Computing, but also anticipating futuristic disruptions on the next-generation integrated HPC-Quantum Computing systems.

Combining Professor Iyer’s probabilistic graphical models and Cao’s real-world measurements have shown to be a success formula in multi-disciplinary domains, from diagnosing diseases to preempting attacks and predicting AI workloads disruption. The team has published deep insights on attack evolutions and error propagation models in top USENIX Security, USENIX Networked Systems Design and Implementation (NSDI), and IEEE Quantum Computing and Engineering (QCE) in collaboration with Associate Professor Gang Wang at Siebel School of Computing and Data Science, Dr. Santiago Nunez-Corrales at the Illinois Quantum Information Science and Technology Center, Anita Nikolich at the School of Information and Sciences. Not only does the team publishes technical findings on attacks and graphics processing unit (GPU) failures in ACM Supercomputing with IBM Research team (Saurabh Jha, Chandra Narayanaswami, Daby Sow, Constantin Adam), Cao also exhibited his artwork of Denial of Service Attacks in Supercomputing’s Art of HPC session, showcased at Georgia World Congress Center and Denver Art Museum.

The insights from investigating security attacks in Cao’s research also translated into diagnosing reliability issues of GPUs in a way that has not previously been done. “GPUs have been thoroughly tested in controlled environments, but their true resilience in real-world applications is still being determined. This is critical given the wide spectrum and rapidly evolving high-performance computing (HPC) workloads, including scientific simulations and the entire lifecycle of AI from training to inference”, Cao said. Using DeltaAI, Cao works with Iyer’s students to create a “data lake” with historical information gathered from errors, disruptions and failures caused by both technical issues. The team will present the finding in the Supercomputing 2025 conference, and will further investigate resiliency in different GPU architectures at exascale.

With NSF Cybersecurity Innovation for Cyberinfrastructure (CICI)’s award, Cao will advance a security data lake “AICyberLake” that provides an evaluation platform for researchers to validate their cyberattacks detection model on live, realistic production network traffic. This research has been built on successful data-driven security approach initiated by Iyer, Zbigniew T. Kalbarczyk, Alex Withers, and Adam Slagell now at the DOE’s Energy Sciences Network (ESnet) and William Kramer’s Blue Waters approximately one Petabyte Data Sets. “For end-users, the crucial thing isn't whether a system is under attack or has failed, but whether their jobs are getting completed with minimal disruptions. That's why our data lake combines operational data on both reliability and security,” Cao said. He plans to work with researchers, e.g., Open Science Data Federation, to disseminate early signals of evolving attacks to prevent them from disrupting NSF-funded advanced computing systems. As a TrustedCI Fellow, at the NSF Cybersecurity Center of Excellence, Cao manages a broad research portfolio addressing crucial security problems including formalizing scientific tokens, measuring quantum-resistant cryptography. He frequently gives tutorials on security log analyses and leads bird of a feather (BoF)’s community sessions at NSF’s major facilities, e.g., the National Center for Atmospheric Research, national labs including Lawrence Berkeley, and agencies such as National Institute of Standards and Technology’s Secure HPC working group. To broaden security findings, he collaborates broadly with academia, national labs, e.g., Argonne Leadership Computing Facility and Oak Ridge National Laboratory, and industry, e.g., Corelight and IBM-Illinois Discovery Accelerator Institute. Some of his mentees have went on and achieved Fiddler Innovation Fellowships for their work on securing supercomputing environments.

By working closely with operators of NSF’s advanced computing systems such as DeltaAI and participating in the Joint Laboratory for Extreme Scale Computing (JLESC), Cao recognized the community’s needs of safeguarding the integrity and security of science, leading to the second NSF award. NSF’s Research on Research Security Program (RoRS) award will support Cao to work with AI/HPC operators to safeguard high performance computing (HPC) resource allocations, which can be applied to flagship programs such as Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) and National Artificial Intelligence Research Resource (NAIRR). This effort will leverage the security data lake above and the massively parallel processing power of GPUs to protect themselves. The expected impact that Cao will outline in the upcoming Academic Security and Counter Exploitation program includes concrete case studies, strengthening research security policies trainings in the research and education sector.

The NSF grants for the two projects reflect Illinois’ exceptional abilities in this area. “Very few people have this systems and AI experience with feet firmly planted on both sides - and it’s not just our group,” Iyer said. “I think Illinois is probably the leader in that.”

Phuong Cao is research scientist / cybersecurity research specialist at the National Center for Supercomputing Applications (NCSA) in the Incident Response and Security (IRST) / Integrated Cyberinfrastructure (ICI) division. He is affiliated with the Center for Artificial Intelligence Innovation. Cao is a TrustedCI Fellow, National Science Foundation Cybersecurity Center of Excellence.

Ravishankar K Iyer is a George and Ann Fisher Distinguished Professor in the Electrical and Computer Engineering Department with affiliations in the Siebel School of Computing and Data Science, Coordinated Science Laboratory, Information Trust Institute, Biomedical and Translational Sciences, Center for Global Studies, Carl R. Woese Institute for Genomic Biology and the National Center for Supercomputing Applications (NCSA).