KREONET

We are thrilled to announce that Mazahir Hussain, a former PhD candidate at the KREONET Center (KISTI), has been awarded the prestigious FABRIC Golden-Stitch Best Experiment Award. Announced at the KNIT12 conference in Spring 2026, this accolade recognizes pioneering experimentation that demonstrates how global research infrastructures can integrate with deeply programmable networks to support next-generation scientific discoveries.

The groundbreaking experimental results that led to this award were published in the paper titled "SENSE in Practice: Quantifying the End-to-End Benefits of Intent-Based Bandwidth Reservation for Exascale Science Workflows"at the SC25 Workshops in November 2025. The study was co-authored alongside international collaborators from Lawrence Berkeley National Laboratory (LBNL), ESnet, the University of California San Diego (UCSD), and the University of Amsterdam.

The Challenge: Beyond "Best-Effort" Networks for Big Data Science

Modern scientific instruments generate astronomical volumes of data. For instance, the Large Hadron Collider (LHC) routinely s petabytes of data across continents every single day. To transfer and process these massive datasets, global collaborations rely on a patchwork of distributed computing facilities and high-capacity research networks that are built and operated completely independently of one another.

Historically, scientific applications have been forced to view the network as an "opaque infrastructure". Researchers inject their high-value data into a simplistic socket interface and "hope" that it arrives at its destination with acceptable performance. Because these applications cannot dynamically negotiate network parameters, scientific workflows frequently suffer from highly variable and unpredictable data transfer speeds.

The Solution: End-to-End Resource Provisioning with Quality-of-Service Guarantees in Heterogeneous Networks

To resolve this bottleneck, the research team evaluated the SENSE (Software-Defined Network for End-to-end Networked Science at the Exascale) paradigm. SENSE introduces an intent-based architecture that fundamentally transforms the network into a "first-class schedulable resource," allowing applications to communicate directly with network management layers to request guaranteed bandwidth and predictable performance.

To rigorously test SENSE under real-world conditions, Mazahir and the team constructed the MIST (Multi-Infrastructure SENSE Testbed). This geographically distributed, multi-domain testbed interconnected three state-of-the-art cyber-infrastructures:

FABRIC: An NSF-funded, deeply programmable national network platform.

The National Research Platform (NRP): A stretched, multi-tenant scientific Kubernetes cluster.

GNA-G AutoGOLE/SENSE*: A global network automation framework.

 *GNA-G AutoGOLE/SENSE WG (Chairs: Tom Lehman/ESnet, Buseung Cho/KISTI, Marcos Felipe Schwarz/RNP, Hans Trompert/SURF)

Quantifying the Benefits: Key Experimental Findings

The MIST testbed allowed researchers to exee full-stack workloads that modeled the active production LHC Compact Muon Solenoid (CMS) infrastructure. The empirical findings presented in the paper successfully quantified the performance advantages of intent-based networking:

1. Rapid and Scalable Network Provisioning

The SENSE orchestrator proved remarkably consistent in translating high-level application intent into automated physical configurations across multiple administrative domains:

Simple Services (CREATE commands): Successfully provisioned guaranteed paths in an average of 183 seconds.

Complex Service Operations (MODIFY commands): Completed multi-domain path modification in an average of 290 seconds.

Workflow Modifications: Control plane actions, such as REPROVISION, averaged 178 seconds, while teardown tasks like CANCELREP and CANCEL achieved swift exeion times of 131and 129 seconds, respectively.

While a configuration overhead exists, the team notes that for exascale workflows and AI training datasets that take hours or days to transfer, an initial setup delay of a few minutes is entirely negligible. The critical benefit is deterministic, guaranteed path stability throughout the transfer.

2. Accelerating High-Priority Scientific Flows

In end-to-end application experiments, the team simulated a massive data transfer scenario involving three separate 12 Terabyte (TB) datasetsrouted between high-performance storage nodes running the production LHC data management stack (incorporating Rucio, FTS3, XRootD, and the Data Movement Manager).

The results demonstrated a stark contrast against traditional methods:

Under standard Best-Effort networking, all data streams competed for the same link capacity, leading to fair but highly prolonged and synchronized completion times as flows congested the backbone.

With SENSE-Provisioned QoS, the network dynamically responded to application-driven priorities. SENSE enforced strict bandwidth separation, allowing high-priority scientific data streams to isolate their traffic and complete significantly fasterthan best-effort alternatives.

A Shared Milestone for Global Open Science

This research marks a significant milestone in realizing the vision of the Department of Energy’s Integrated Research Infrastructure. By bridging the gap between application requirements and multi-domain network orchestration, this work paves the way for predictable, secure, and automated high-throughput data spaces for international scientific collaborations.

The project received support from the U.S. Department of Energy (LBNL), the US National Science Foundation (NSF), Ciena, and the Korea Institute of Science and Technology Information (KISTI) (Grant No. K25L5M1C1). In alignment with open science principles, all benchmarking configurations, collection materials, and reproducible scripts have been open-sourced and are publicly accessible via the SENSE Performance GitHub Repository.