Data is a key resource for research in networked systems, from evaluating systems or protocols to training new AI/ML models. However, real-world data is often unavailable or unusable (e.g., due to privacy regulations), and replicating production networks exactly is expensive or even impossible. Testbeds and digital twins offer a controlled alternative for generating realistic datasets, provided the experiments behind them are reproducible and scalable.

This talk discusses the challenges of generating research data with testbeds and presents approaches to address them. We first introduce the plain orchestrating service (pos), a testbed framework that makes experiments reproducible through a well-defined, fully automated workflow. We then address scalable dataset generation: HVNet uses hardware-accelerated virtualized networks on a single physical host with real NICs and cables, yielding more realistic latency behavior than common emulation tools. Use cases such as latency modeling with Graph Neural Networks (GNNs) and digital twins for fault management illustrate how such datasets can be created and used. Looking ahead, we discuss how an Internet Digital Twin, a measurement-based virtual clone of the Internet, could enable what-if experiments at scale. We conclude by showing how these building blocks come together in SLICES-DE and SLICES-RI, which connect distributed testbeds into a large-scale research infrastructure with shared reference experiments (blueprints), long-term availability, and FAIR-aligned data management.

About Johannes Späth

Johannes Späth received his B.Sc. (2021) and M.Sc. (2023) degrees in Informatics from the Technical University of Munich (TUM). He is currently pursuing a Ph.D. at the TUM Chair of Network Architectures and Services. His research interests include data-driven network management, digital twin systems, autonomous networks, and the QUIC protocol.

This event will be conducted in English