The quest for predictable and secure networking for science

The exponential growth of data from scientific instruments, distributed computing, and artificial intelligence workflows has strained global cyberinfrastructure. Compute, storage, and the network must scale together to meet exascale demands and rapidly changing workloads. The network remains a passive, best-effort transport substrate that offers applications no interactive negotiation, multi-domain coordination, or predictable performance. This thesis addresses the gap between the deterministic requirements of distributed scientific workflows and the unpredictability of traditional networking by introducing the Software-Defined Network for End-to-End Networked Science at the Exascale (SENSE) architecture, which elevates the network to a first-class, schedulable resource.
SENSE combines intent-based networking with full-stack semantic resource modeling. Using the Multi-Resource Modeling Language (MRML) together with the Network Markup Language (NML) and the Network Services Interface (NSI), it establishes a hierarchical orchestration framework supporting interactive, bi-directional negotiation in the control plane. Applications translate high-level intents, such as deadlines and dataset sizes, into provisioning directives that structurally enforce the security policies of autonomous domains and coordinate state from the wide-area backbone to the host stack inside Science DMZs, resolving the "Last Inch" problem.
The architecture is validated operationally through prototype integration with production pipelines, including the LHC/CMS Rucio/Data Movement Manager and near-real-time ExaFEL analysis, and empirically on the Multi-Infrastructure SENSE Testbed (MIST), federating FABRIC, NRP, and AutoGOLE resources. Results show that dynamic application-to-network orchestration reduces completion-time variability for multi-terabyte elephant flows relative to best-effort routing. The framework is being tested and deployed on research and education networks worldwide.