HPC

This talk outlines a holistic framework for network digital twins (NDTs) in high-performance computing, covering construction, optimization, and application. The approach integrates federated learning for privacy-preserving model updates and reinforcement learning for closed-loop control, enabling real-time adaptation to traffic and system dynamics. Topics Covered HPC network digital twin construction: topology and telemetry ingestion, calibration Distributed and regional twin orchestration Federated learning for privacy-preserving NDT updates Reinforcement learning for closed-loop HPC network control Case studies: workload forecasting, congestion-aware routing, anomaly detection Portability to edge and 6G-class environments Links Workshop Program

Oak Ridge National Laboratory (ORNL) Duration: [Start Date] – [End Date] (Please update the dates above as needed.) Project: ExaDIGIT — Exascale Digital Twins for HPC Network Systems Focus Areas: Design and development of exascale-level network digital twin frameworks Physical-digital synchronization for high-performance computing fabrics Topology-aware network representation and performance modeling Integration with ORNL’s production HPC systems Recognition: ExaDIGIT project honored with the 2025 R&D 100 Award