KREONET

Every second, the detectors of the Large Hadron Collider (LHC) at CERN record more than a billion proton-proton collisions. After real-time filtering, the four main experiments—ATLAS, CMS, ALICE, and LHCb—still generate tens of petabytes of data annually. When the accelerator is upgraded to its high-luminosity phase later this decade, this volume is expected to reach into the hundreds of petabytes. Moving these datasets from the French-Swiss border to analysis centers across five continents requires far more than just bandwidth; it demands networks that can be provisioned, monitored, and adapted similarly to how large experimental facilities manage their detectors and accelerators.

This is precisely the goal of SENSE (Software-defined networking for science at exascale), an orchestration framework developed at the U.S. Department of Energy’s Energy Sciences Network (ESnet), in collaboration with several partners. SENSE considers the network as a first-class scientific instrument, capable of reserving capacity, satisfying strict quality-of-service requirements, and offering researchers “intent-based” services—requests phrased in terms of science rather than hardware.

Mazahir Hussain, a Ph.D. candidate at KISTI’s KREONET in South Korea, has spent the past few months at Berkeley Lab as an ESnet visiting researcher, working on performance evaluation of the SENSE project for the data-intensive era of the High-Luminosity LHC. Working with Justas Balcas, Xi Yang, and ESnet director Inder Monga, Mazahir helped develop MIST—the Multi-Infrastructure SENSE Testbed—which connects three different research cyberinfrastructures:

• FABRIC, the U.S. National Science Foundation’s adaptive, programmable network platform

• The National Research Platform (NRP) based at the Supercomputer Center of UC San Diego

• AutoGOLE, an international testbed operated by research and education networks, where KREONET provides two data-transfer nodes via the Network Service Interface (NSI) protocol

MIST currently links six FABRIC sites (CERN, AMST, NEWY, STAR, UCSD, LOSA), Caltech on AutoGOLE, and UCSD on the NRP. It manages them with workflows patterned after the CMS experiment’s global data-handling model. The design is based on three principles: 1) heterogeneity—multiple administrative domains and hardware types; 2) comprehensive, reproducible instrumentation; 3) full-stack, resource-intensive workloads that mirror real scientific processes.

Photo: Mazahir Hussain with Chin Guok, Chief Technical Officer of ESnet, at the Poster Session at Berkeley Lab

Results show that SENSE can establish end-to-end circuits within minutes in multi-domain, sustain line-rate transfers amid competing traffic, and provide detailed telemetry. Mazahir shared these findings at Berkeley Lab’s recent Scientific Computing Poster Session, demonstrating how dynamic network orchestration can enhance data transfer performance for LHC experiments by accommodating different priorities.

This internship strengthens the connection between KREONET and ESnet at a time when global science is gearing up for unprecedented data production. By proving that networks can be programmed as confidently as supercomputers and detectors, the MIST project brings the community closer to a future where transferring a petabyte becomes as routine as scheduling a job on a cluster—exactly what the next decade of LHC research will demand.