Accepted NRE Demos

SC26-NRE-101 PDF

The setup will demonstrate the feasibility of deploying secure connectivity solutions and networks that implement a security-in-depth approach by providing the best connectivity solution taking into account the underlying key distribution solution which is available between network locations: Quantum Key Distribution (QKD) or Post Quantum Cryptography (PQC).

This is possible by using the QuLore framework developed by EHU and two R&E developed KMS: edukms developed by SURF and vMKS by EHU. We will demonstrate Quantum Safe communications at three different layers: at the network layer between two enterprise networks using IPsec, at the optical layer over DWDM systems and at the application layer finally between end user applications.

The demonstration will involve several links:

• Two pairs of DWDM systems deployed between CIENA and PCCS booths at the SuperComputing Exhibition Floor and between two locations in the University of the Basque Country Bizkaia campus: The Faculty of Engineering of Bilbao and the Advanced Aeronautical Manufacturing Center (CFAA) in Zamudio. These links will co-propagate quantum and data channels over a shared fiber.
• Two IPsec-secured links with PQC cryptographic protocols between PCSS booth and SURF location in Amsterdam and between this same place and CFAA in Zamudio.
• IP links between different end-user applications. In this case, an Open-Source videoconferencing application has been modified to make use of quantum-safe keys for encrypting the audio and video traffic. These keys can be both direct QKD or relayed QKD using PQC keys for the relay, depending on the endpoints.

The demonstration brings together Ciena’s Waveserver quantum-safe encryption and LuxQuanta’s QKD technology featuring a WaveLogic 6 Extreme (WL6e) 1.6Tb/s encrypted channel in Chicago and a WaveLogic 5 Extreme (WL5e) 800Gb/s encrypted channel in Bilbao. Both locations showcase a hybrid quantum-safe security architecture, combining NIST-certified PQC algorithms implemented in the Waveserver optical encryption solution, with QKD-generated keys from LuxQuanta NOVA LQ systems, to demonstrate how complementary quantum-safe technologies can be integrated with proven high-capacity optical encryption to ensure the confidentiality of critical in-flight data. Two QRNG will provide true random numbers for PQC Key Establishment Mechanism (KEM) secure implementation. 

SC26-NRE-102 PDF

perfSONAR is widely used to measure R&E networks, with OWAMP (One-Way Active Measurement Protocol) providing per-direction latency visibility that round-trip tools like ping cannot. But accurate one-way latency depends on tightly synchronized clocks at both endpoints, and when nodes are topologically close (intra-campus, intra-metro, or within a single regional network), the real latency often falls inside the error envelope of NTP synchronization. The result is a familiar artifact: apparent negative one-way latencies and genuine path asymmetries hidden in the noise.

This NRE brings two regional networks into SCinet’s Xnet precision timing infrastructure at SC26 to tackle exactly that problem. Metropolitan Research & Education Network (MREN) primarily serves universities and national labs across seven states in the upper Midwest, anchored at the StarLight exchange facility in Chicago. Capital Area Advanced Research & Education Network (CAAREN), run by George Washington University, is the DC area’s on-ramp to Internet2, connecting R&E institutions across DC, Northern Virginia, and Maryland. Both networks run perfSONAR today and want to improve those measurements by giving their hosts access to better time.

SC26-NRE-103 PDF

The SKA Observatory (SKAO) is an international organization responsible for constructing and operating the Square Kilometre Array (SKA), one of the world’s largest radio astronomy projects. The SKA telescopes will generate unprecedented volumes of observational data that must be reliably transferred between SKA Regional Centres (SRCs) distributed around the world. Within the SKA Regional Centre Network (SRCNet), high-throughput and stable data transfer capabilities are essential for efficient data management and scientific research.

NEC’s Ultra-high-speed Data Transfer System (UDTS) has demonstrated high-speed disk-to-disk transfer performance over 100 Gbps-class networks. Through a proof-of-concept (PoC) study, we are evaluating the integration of UDTS with the existing SRCNet software stack, consisting of SKA IAM, Rucio, FTS, and XRootD, to accelerate inter-site data transfers while preserving the existing data management and transfer control workflows.

At SC26, we conduct a live demonstration of an SRCNet-compatible data transfer system connecting three sites in Japan and the United States. The demonstration verifies end-to-end interoperability between the SRCNet software stack and UDTS and evaluates UDTS’s transfer performance over a 100 Gbps-class network using datasets of varying file sizes and file counts, including Science Data Challenge (SDC) datasets.

SC26-NRE-105 PDF

This demonstration presents an international 8K uncompressed video processing workflow over a 400 Gbps network, enabled by SRv6-based in-network computing. We use SRv6 End.AN to integrate packet forwarding with computational processing in the network data plane. By processing media along the forwarding path, the approach aims to avoid the traffic hairpinning associated with conventional NFV-based service chaining. High-speed forwarding and SRv6 processing are supported by our open-source, DPDK-based SRv6 software router.

At SC26, we demonstrate this architecture across a network connecting Japan and the United States. Multiple video processing functions (VVFs) for 8K uncompressed streams are dynamically chained on demand to create end-to-end media workflows. A custom-developed monitoring system capable of handling 400 Gbps traffic visualizes transmission conditions and resource utilization in real time. The demonstration explores how tightly integrated networking and computing resources can support next-generation, ultra-high-bandwidth media applications

SC26-NRE-106 PDF

The National Research Platform (NRP) AI Ecosystem provides a federated, Kubernetes-native environment for LLM inference and serving, hosting a rotating catalog of open-weights LLMs accessible via an OpenAI-compatible API and Envoy AI Gateway across 80+ member institutions.

At SC26 we extend this ecosystem to the wide-area network by orchestrating dynamic, QoS-enabled optical Layer 2 paths between three distributed anchor points: the SCinet cluster on the exhibit floor, the Prototype National Research Platform (PNRP) distributed across SDSC, MGHPCC, and UNL, and Internet2’s national backbone. Using the ESnet SENSE Orchestrator and the NRP’s native SENSE Kubernetes Operator, this NRE provisions deterministic, high-bandwidth connectivity for live, multi-site LLM inference and agentic multi-turn workflows: cross-site generation, retrieval-augmented generation (RAG) over a shared embedding store, and pipeline-parallel experiments that measure how WAN latency shapes end-to-end serving performance. By integrating SCinet into the NRP’s coast-to-coast AI research fabric, attendees interact with models that physically span three administrative domains.

SC26-NRE-107 PDF

The convergence of programmable data planes and Kubernetes has long been hampered by an asymmetry: while compute and storage are declarative resources (Pod, PersistentVolumeClaim), guaranteed-bandwidth WAN circuits and P4-programmable hardware have remained outside the Kubernetes control plane. At SC26 we close this gap with a fully Kubernetes-native P4 testbed that treats network paths as first-class resources. The SENSE Operator (a production Kubernetes operator elevating multi-domain L2 circuit provisioning to a SensePath CRD) dynamically provisions deterministic, QoS-enabled optical paths from SCinet to a geographically distributed P4 infrastructure (SmartNICs, BlueField-2 DPUs, Tofino2 switch) integrated into the NRP Nautilus cluster. Two novelties: (1) a Kubernetes-native network resource abstraction via the operator’s declarative reconcile loop, and (2) a production P4 testbed living entirely inside Kubernetes, with programmable data planes accessible via kubectl, from SmartNIC bitstream deployment to live packet telemetry.

SC26-NRE-108 PDF

The San Diego Supercomputer Center and the Prototype National Research Platform (NSF Award 2112167, Category II) present the NRP, a community-owned federation of over 80 institutions. At SC26, NRP’s Nautilus Kubernetes Cluster directly integrates SCinet-connected compute nodes on the exhibit floor for live, interactive AI/ML and HPC workflows. Beyond the local demonstration, this NRE conducts wide-area network experiments with the ESnet SENSE Orchestrator and NRP’s SENSE Operator, provisioning on-demand, high-bandwidth, deterministic QoS paths across SCinet, StarLight, FABRIC, Caltech, and the Chameleon Cloud, transforming the show floor into a national-scale research instrument.

SC26-NRE-109 PDF

International collaboration drives modern science projects, demanding rapid data transfer across borders. To support this, academic networks are enhancing international lines. In April 2025, SINET6, a Japanese NREN, launched the first 400 Gbps international line across the Pacific Ocean. This experiment is the world’s first data transfer trial between Japan and the USA to utilize a network configured with end-to-end 400G technology.

MMCFTP (Massively Multi-Connections File Transfer Protocol), developed by NII, is a powerful data transfer tool. In previous SCs, MMCFTP demonstrated transfers using multiple 100 Gbps lines. These experiments involved memory-to-memory data transfer and were not limited by server file I/O performance. The SC26 experiment will demonstrate disk-to-disk file transfers over a single 400 Gbps end-to-end line.

SC26-NRE-110 PDF

To investigate the programmable network technology over large-scale topology, we integrate P4 switching and computing resources of participating parties through several domestic and international circuits. In order to implement smart network control, we exploit Intent-Based Networking (IBN) technology which translates high-level business objectives into actionable network policies. Using natural language processing and semantic reasoning, the IBN interprets user intent (from formal policy languages to natural language), resolves ambiguities, handles conflicts, and decomposes complex intents into atomic operations mappable to specific configurations and routing decisions. We will demonstrate practical feasibility of multi-domain smart network control using IBN through international collaboration, contributing both academic knowledge and practical evidence for next-generation network infrastructures.

SC26-NRE-111 PDF

The research team plans a first-of-its-kind demonstration of deployed quantum-classical coexistence with quantum entanglement distribution at SC26. The proposed NRE seeks to demonstrate co-existence of quantum and classical communications on a multi-node network over deployed commercial dark fiber spanning 35 kilometers between Northwestern University in Evanston, the International Center for Advanced Internet Research (iCAIR)/StarLight facility, and the SC26 venue in downtown Chicago. This effort, led by Professor Prem Kumar and Gina Talcott, is focused on designing entangled photon sources and quantum network protocols that enable quantum measurements between physically separated locations, moving entanglement-based coexistence experimentation beyond current laboratory or loopback limitations. Quantum photons will transmit alongside SC26 conference infrastructure traffic spanning the C+L-band, as well as an S-band synchronization signal to enable GPS-disciplined synchronization and picosecond-level syntonization. Furthermore, a quantum network orchestrator will employ software-defined networking principles to provide end-to-end control of the network, autonomously coordinating all calibration and quantum measurements. This NRE will significantly advance implementations of hybrid quantum-classical networking for real-world network environments.

SC26-NRE-112 PDF

This paper presents a field-programmable gate array (FPGA)-based SmartNIC architecture designed to address critical network challenges in data-intensive scientific computing. Our approach focuses on implementing network visibility, optimization, and orchestration capabilities through FPGA acceleration. The goals are to develop a flexible firmware-centric implementation of high-bandwidth network functions on AMD/Xilinx SmartNICs, leveraging P4-based packet-processing methods with support for packet parsing, marking, telemetry, local steering, and control hooks for machine learning inference.

The next generation of data-intensive science will be constrained by networks as much as by compute and storage. Major computing sites are preparing for sustained wide-area traffic in the 400 Gb/s region and beyond. Visibility is needed to identify the owner and purpose of traffic; optimization is needed to make more efficient use of shared bandwidth; and orchestration is needed so workflows can request elevated service when required.

As a concrete example, we have implemented and evaluated a preliminary network telemetry architecture based on Scientific Network Tags (SciTags). SciTags associate network traffic with scientific context through packet marking. The FPGA-based bridge counts IPv6 packets according to the experiment information in their SciTag values, carried in the IPv6 Flow Label, and performs programmable virtual local area network (VLAN) translation. Flowd marks packets on the external nodes. Host software configures the tables and periodically reads the counters into a local database. The setup was implemented and evaluated on AMD Alveo U280 FPGAs in the Open Cloud Testbed (OCT) at the Massachusetts High Performance Compute Center (MGHPCC).

Using iperf3, we compared the FPGA-based bridge with a reference path between regular network-interface cards (NICs). Receiver throughput was 37.0 Gb/s for both paths on 40 Gb/s NICs, and 95.1 Gb/s for the reference versus 95.3 Gb/s for the bridge on 100 Gb/s NICs. These results show no observable throughput degradation. The FPGA counters agreed exactly with offline packet-capture analysis: ATLAS, 854,311 packets; CMS, 744,050; LHCb, 765,889; and ALICE, 841,103.

SC26-NRE-113 PDF

This demonstration will build on our previous NRE demonstrations with significant advancements. We aim to show intelligent, dynamic arrangement and re-arrangement of widely distributed processing of large volumes of data across a set of compute and network resources organized in response to resource availability and changing application demands. We also aim to explore performance limitations and enablers for high volume bulk data transfers. Our network will be assembled using a number of switches and multiple SCinet Tbps connections. We plan to show rapid automated deployment and redeployment with prioritized optimization, real-time monitoring and QOS management of application data flows with very different network demands. Technologies we intend to leverage include SDN, RDMA, RoCE, NVMe, GPU acceleration and others.

SC26-NRE-114 PDF

This demonstration will aim to deliver multiples of the bandwidth that was available to NRE demonstrations in previous years. The focus is exploring and exploiting the multiplied bandwidth newly available to enterprise networks through the breakout availability of Tbps coherent communications channels that can ride open optical line systems straight through to the enterprise data center network. These coherent channels are able to bypass the cost and complexity of intermediate networks and land dramatic network bandwidth directly in the data center and direct access to spectrum services provides other important value to certain missions.

SC26-NRE-115 PDF

Cosmic rays continuously pass through Earth, carrying information about high-energy astrophysical processes such as supernovae, active galactic nuclei, and other extreme environments. Detecting scientifically meaningful events at scale is difficult. Candidate signals are rare, observations are noisy, and detector nodes may be geographically distributed across universities, schools, citizen-science participants, and edge platforms. Moving all raw data to a central site is inefficient and may not scale as detector networks grow.

We propose an SC26 Network Research Exhibition demonstrating a distributed cosmic ray sensor network that combines low-cost CosmicWatch muon detectors, smartphone-based data from the Cosmic-Ray Extremely Distributed Observatory (CREDO), real-time data streaming, and federated/edge machine learning. The demonstration will use SCinet and connected research infrastructure to demonstrate how federated/edge learning can make distributed scientific instrumentation more scalable by reducing raw-data movement while still improving a shared model across sites.

SC26-NRE-116 PDF

With multiple national and international partners, iCAIR is designing, developing, implementing, and operating an international Software Defined Exchange (SDX) at the StarLight International/National Communications Exchange Facility, which integrates multiple services designed for large scale global data intensive science. The StarLight SDX is based on a flexible, scalable, programmable platform. This SDX, which is managed by a multi-organizational consortium, has been proven able to integrate many different multi-domain services and to ensure services isolation, e.g, to segment large capacity science flows from other flows, and to segment research testbed flows from production flows, including segments for connecting quantum networking testbeds, which will be supported at SC26. Services include those based on Gen5 400 Gbps Data Transfer Nodes (DTNs) for Wide Area Networks (WANs), including transoceanic WANs. Currently, a key focus is scaling to multi-400 Gbps, 800 Gbps, 1.2 Tbps, 1.6 Tbps WAN and LAN E2E technologies that provide high performance transport services for exascale science, controlled using Software Defined Networking (SDN) and AI techniques. Planning has begun on evaluating Gen6 DTNs and 800 Gbps switches. SDN enabled DTN services are being designed specifically to optimize capabilities for supporting large scale, high capacity, high performance, reliable, high quality, sustained individual data streams for science research.

SC26-NRE-117 PDF

The Global Research Platform (GRP) is an international scientific collaboration that is creating innovative advanced ubiquitous services that integrate resources around the globe at speeds of multi-gigabits and terabits per second, especially for data-intensive science research. GRP focuses on design, implementation, and operation strategies for next-generation distributed services and infrastructure to facilitate high-performance data gathering, analytics, transport, computing, and storage among multiple science sites at 100 Gbps or higher (e.g., 400 and 800 Gbps WAN streams as well at Tbps). GRP community partners are located in North America, Asia, Europe, and South America and work together to customize international fabrics and distributed cyberinfrastructure to support optimal data-intensive scientific workflows. Essentially, the GRP is a worldwide Science DMZ, a distributed environment for data-intensive research. The GRP leverages optical circuits and open exchange facilities provided by its collaborators. The GRP consortium focuses on extremely high-capacity data intensive science. Some facilities used are dedicated to specific science projects, some are shared under restrictive policies with other science communities, others are more widely shared with research communities. The GRP assists in coordinating among those activities.

The GRP focuses on recent and emerging advances in architecture, services, technologies, and infrastructure, specifically recent innovations that can directly support global large-scale data-intensive science research via: (a) Large Scale Global Science; (b) Next-Generation Research Platforms; (c) Orchestration Among Multiple Domains; (d) Large-Scale Data WAN Transport; (e) High-Fidelity Data Flow Monitoring, Visualization, Analytics, Diagnostics, Event Correlation, AI/ML/DL; (f) Data-Intensive Science, Programmable Networking, and Automation; and, (g) International Testbeds for Data-Intensive Science.

Each year, the Global Research Platform community organizes an annual workshop to present their innovations to wider communities. This workshop is co-located with the IEEE International Conference on eScience. The 2024 GRP/eScience event took place in September in Osaka, Japan. The 2025 workshop took place in September in Chicago, Illinois. The 2026 GRP workshop will be held in Naples, Italy. In addition, GRP workshops have been held co-located with the Supercomputing Asia Conference in Singapore and the National Research Platform workshop.

SC26-NRE-118 PDF

This submission has no standalone abstract, so the opening of its Executive Summary is used below.

The Global Network Advancement Group (GNA-G), through its Data-Intensive Sciences (DIS) and SENSE/AutoGOLE Working Groups, is developing a next-generation, multi-terabit/sec global system to support the rapidly escalating demands of data-intensive science. Scientific programs such as the LHC, Rubin Observatory LSST, SKA, Cryo-EM, genomics, and distributed AI/ML workflows now generate unprecedented data volumes, requiring coordinated global infrastructure capable of predictable, high-throughput, multi-domain operation.

Building on the progress of previous Supercomputing conferences, especially SC25 and further developments since then, this NRE will deploy a state of the art set of end-systems capable of multi-Tbps throughput supporting AI-integrated workflows through many partner NREs in a wide range of disciplines.

This system integrates programmable networks (P4, SRv6, PolKA, BGP-CT), multi-domain orchestration (SENSE/AutoGOLE), real-time telemetry (INT, perfSONAR, ALTO, GradientGraph), ML-driven intelligence (NetPredict, Hecate, GradientGraph), high-throughput end systems (400G to 800G DTNs, GPU nodes), and regional caches/data lakes into a unified, composable architecture. It is validated through global testbeds such as GP4L, FABRIC, NRP, and GRP, and deployed across multi-Tbps optical infrastructures including APOnet, AmLight, Rednesp/RNP, ESnet, Internet2, GEANT, and StarLight.

At SC26, this NRE and more than a dozen partners NREs will demonstrate this architecture at scale, showcasing multi-domain orchestration, programmable data planes, AI-driven optimization, Digital Twin simulation, and multi-Tbps global data movement. The system supports a progressive deployment model spanning R&D, advanced production, and full production, ensuring coexistence with general R&E traffic while enabling the next decade of scientific discovery.

SC26-NRE-119 PDF

Over the past decade, cryo-electron microscopy (Cryo-EM) has emerged as an important technique in structural biology, dramatically expanding our capacity to visualize biological macromolecules at near-atomic resolution. This advancement has been particularly impactful in biomedical research, enabling scientists to determine the detailed structures of critical targets such as the HIV vaccine research and SARS-CoV-2 spike protein. Insights obtained from Cryo-EM have directly accelerated vaccine and therapeutic development, demonstrating its potential to transform our approach to tackling infectious diseases, many of which disproportionately burden the Global South. Despite these advances, researchers in resource-constrained settings, particularly in Sub-Saharan Africa and other Global South locations, face significant barriers in accessing Cryo-EM infrastructure due to the high cost, sophisticated instrumentation, and specialized technical expertise required.

To address these access and infrastructure constraints, we extend FABRIC to South Africa, introducing a transcontinental testbed that brings advanced networking capabilities closer to under-resourced regions while exposing the practical limits of latency, data movement, and distributed coordination at intercontinental scale. We then apply FABRIC Loom AI to explore network-embedded single-particle (SPA) Cryo-EM processing through selective in-network placement of Cryo-EM data preprocessing stages, reducing reliance on centralized facilities and enabling more accessible, distributed analysis workflows.

SC26-NRE-120 PDF

Large scale data production within and among science research collaborations and sites continues to increase, a long-term trend that continues to accelerate rapidly, driven especially by the deployment of new science facilities, including high luminosity research instrumentation and next generation computational science centers. Consequently, the science networking community is planning networking infrastructure that can support services beyond 100Gbps, with one thematic focus on 400 Gbps LANs and WANs (represented also by other NRE demonstrations planned by this consortium). However, this consortium is also investigating capabilities for WAN services beyond 400 Gbps, including 800 Gbps, 1.2 Tbps, 1.6 Tbps, and multi-Tbps WAN and LAN services. This NRE describes the required infrastructure for prototype multi-Tbps WANs. NB: A companion NRE describes a demonstration of the Multi-Tbps services that will be supported by this infrastructure.

SC26-NRE-121 PDF

Large scale data production within and among science research collaborations and sites continues to increase, a long term trend that continues to accelerate, especially driven by the deployment of new science facilities, including high luminosity research instrumentation and next generation high performance computation science centers (many AI centric). Consequently, the science networking community has begun to prepare infrastructure that can support high-capacity services, with a key thematic focus on services beyond 400 Gbps LANs and WANs, including those approaching 800 Gbps, 1.2 Tbps, 1.6 Tbps, and multi-Tbps WAN and LAN services. NB: A companion NRE describes demonstrations of the multi-Tbps infrastructure required for a large scale multi-Tbps services. This NRE describes prototype multi-Tbps WAN services that can be supported by the infrastructure described in that NRE. Also, Ref NRI 103.

SC26-NRE-122 PDF

Domain science workflows often require integration across a heterogeneous set of data, compute, storage, and instruments which are located at different facilities. Operation within this distributed Cyberinfrastructure context is rapidly increasing due to the emergence of new Artificial Intelligence (AI) motivated workflows, the Department of Energy focus on building a virtual “SuperFacility” based on seamless integration across its many facilities, and general technology trends which require dynamic and configurable access to many different types of resources. SENSE is an Orchestration system which enables the instantiation of end-to-end network services which can be customized and presented to workflow agents in application specific context, and includes full lifecycle service modifications, monitoring, and troubleshooting.

The SENSE Orchestration focus is to enable the construction of topologies which integrate distributed Compute, Storage, Instrument, and Network resources for domain science workflows. These types of topologies requires integration across disparate technology domains, administrative boundaries, identity verification, and resource access authorization.

In this context, the SENSE system focuses primarily on the network domain. While all of the technology domains need to know enough about the other integrated domains to access their services, the network component is the “glue” that can bring all of the components together in a functional manner. As a result, SENSE typically has a more detailed understanding and feature sets which extend into the other technology domains. SENSE enables workflow agents to consider network services as part of overall resource planning and scheduling activities.

For instance, SENSE can build an end-to-end guaranteed bandwidth network service which connects Data Transfer Nodes (DTN) or Kubernetes clusters at different sites. In this context end-to-end includes the networks between these two sites, and the network stacks inside the end-systems.

This NRE demonstration will show the SENSE system in action on integrated topologies which include external facilities and also on the SCinet National Research Platform (NRP) Kubernetes Cluster in SCinet NOC. The demonstrated functionality will include:

• Rucio’s use of SENSE services spanning production and prototype systems for LHC CMS and ATLAS facilities at NET2 MGHPCC/UMass, FermiLab, CERN, SCinet NRP Cluster, and others.
• SENSE managed fine grained traffic engineering to adjust bandwidth guarantees in realtime.
• SENSE services integration with prototype LHC site data movement end systems including the site XRootD/dCache clusters.
• Show the SENSE Priority Network service which can be used as part of the LHC Rucio driven data management and movement workflows.
• SENSE Service Specific end-to-end data flow monitoring and troubleshooting dashboard.

SC26-NRE-123 PDF

FABRIC (FABRIC is Adaptive ProgrammaBle Research Infrastructure for Computer Science and Science Applications) is an International infrastructure that enables cutting-edge experimentation and research at-scale in the areas of networking, cybersecurity, distributed computing, storage, virtual reality, 5G, machine learning, and science applications.

The FABRIC infrastructure includes a distributed set of equipment at commercial colocation spaces, national labs and campuses. Each of the FABRIC sites has large amounts of compute and storage, interconnected by high speed, dedicated optical links. It also connects to specialized testbeds (5G/IoT PAWR, NSF Clouds), the Internet and high-performance computing facilities to create a rich environment for a wide variety of experimental activities. The FABRIC Network includes 35 sites distributed across the United States, CERN (Switzerland), University of Amsterdam (Netherlands), University of Bristol (England), and University of Tokyo (Japan).

This demonstration will show the FABRIC Research Cyberinfrastructure in operation with a focus on the following capabilities:

• Experiment specific topology (i.e. slice) instantiation consisting of network embedded compute and customized storage resources. FABRIC compute capability includes CPU, FPGA, GPU, and DPU based resources. FABRIC programmable networking includes automated network provisioning, P4 controlled elements, and realtime monitoring.
• Integration of external resources into FABRIC slices via use of FABRIC “Facility Ports”.
• R&E and commodity Internet routed infrastructure, and production Science DMZs at universities and domain science specific research production research facilities can also be included in FABRIC experiment topologies.

SC26-NRE-124 PDF

The emergence of the Innovative Optical and Wireless Network (IOWN) addresses the growing demand for ultra-low latency, high bandwidth and energy efficient optical networks. While operated in the fundamental level of the transmission network, traffic monitoring and analysis from a higher level are mandatory to better understand the intents and purposes behind the usage. As one of the best candidates, Scientific network tags (scitags) is a technology to label the traffic from the source level. The label can be later collected and analyzed from any point along the transmission path to accurately summarize the traffic from the network flow perspective. Together with the power of the active network measurement and visualization, such as perfSONAR and Grafana, the insight of the network traffic over low level transmission services such as IOWN can be discovered.

In SC26, the feasibility of using scitags and measurement tools on IOWN-inspired networks will be demonstrated, highlighting the potential of this approach for future network architectures.

SC26-NRE-125 PDF

Managing large-scale scientific workflows over international WANs will become increasingly complex as multiple science projects share underlying resources while governed by divergent variables, constraints, and technologies. This NRE demonstration will showcase scientific network tags (Scitags), an open-source initiative promoting high-fidelity visibility into how individual science domains utilize end-to-end network resources. Building upon previous international demonstrations, this SC26 NRE will expand the scope to showcase two distinct areas: advanced 1.6 Tbps networking capabilities and active production workflow monitoring.

The high-rate portion of the NRE will consist of multiple sites sending marked transfers to the Starlight booth at the exhibition floor. The participating sites are the University of Victoria (UVic), CERN, and Fermilab (FNAL), which will be connected at 1.6 Tbps, 1.2 Tbps, and 400 Gbps, respectively. All connectivity will come to SCinet via Starlight. The path from UVic is partly provisioned as two 800 Gbps links using coherent optics; the path from CERN to UVic consists of multiple 300 Gbps paths, provisioned across multiple 400G transatlantic circuits using SENSE.

Each sending site is equipped with multiple 400 Gbps Data Transfer Nodes (DTNs). Transfers will be done with iperf3 to assess the maximum achievable bandwidth, and also with xRootd and dCache to demonstrate the viability of packet marking in disk-to-disk transfers and assess the bandwidth achievable in that case.

We will also conduct a live production demonstration of the flow labelling (“fireflies”), in collaboration with US/ATLAS and US/CMS sites, including Brookhaven National Laboratory (BNL) and Fermi National Accelerator Laboratory (FNAL). The sites will actively participate by exchanging data over production networks. We will produce monitoring dashboards showing the flows sorted by applied scientific activities and will report the live results directly at the StarLight booth on the SC26 show floor. This will demonstrate how any intermediate network or end-site can easily identify both the owner and the purpose of the traffic, providing standardized and detailed visibility into the nature of HEP traffic running to and from their sites.

SC26-NRE-126 PDF

This demonstration presents two high-speed packet processing applications leveraging FPGA-based hardware acceleration for network anomaly detection and financial trading risk management. It builds on our SC25 Network Research Exhibition demonstration, which showcased a hybrid FPGA-P4 architecture for 400G line-rate packet processing. A Tofino2 P4 switch extracts the 5-tuple (source and destination Internet Protocol (IP) address, source and destination port, and protocol) from each incoming packet and buffers it in on-chip registers. After every four packets, the switch packs all four 5-tuples into a single Ethernet frame and forwards it to the FPGA. We extend the same packet-header processing pipeline to two latency-critical applications.

In anomaly detection, the flow-size distribution (FSD) describes how many traffic flows carry a given number of packets and captures the fine-grained structure of network traffic. Small flows, commonly called “mice flows,” account for up to 99% of the FSD. Reconnaissance attacks often consist of such low-packet-count flows aimed at many hosts or ports, so a change in the mice-flow distribution can signal an emerging attack or anomaly. We propose MiceRadar, a compact fingerprint-counter data structure that accurately estimates the flow-size distribution by targeting the mice-flow region. To detect attacks, the system compares the FSD of the current interval with that of the previous interval using the Jensen-Shannon divergence (JSD), a measure of the difference between two distributions; a large JSD indicates a sudden change in traffic. An exponentially weighted moving average (EWMA) keeps a running mean and variance of the JSD, and an anomaly is flagged when the current JSD exceeds the resulting threshold. The threshold adapts to normal traffic changes that reduces false alarms while detecting both high-volume and low-rate attacks.

In financial trading, an order cannot be withdrawn once it reaches the exchange. There’s a need for pre-trade risk checks before orders leave the broker. In collaboration with KGI Securities, we implement pre-trading risk checks on the FPGA. Implemented in the order path between the broker’s trading system and the exchange. The system takes four inputs: Financial Information eXchange (FIX) orders, orders in the Taiwan Futures Exchange (TAIFEX) TMP format, execution reports, and market data. Its five-stage pipeline (1) parses network headers (layers 2 to 4) together with FIX, TMP and market-data messages; (2) reorders out-of-order and retransmitted Transmission Control Protocol (TCP) segments to reconstruct a continuous byte stream; (3) rebuilds FIX messages by identifying message boundaries while keeping parsing state across segments; (4) enforces limits on order amount and quantity, per-instrument positions, order rates, consecutive cancellations and whitelist membership; and (5) maintains a client-level order book in on-chip memory, calculates margins according to product-specific rules, and tracks execution reports and the trade book. A host computer sets the rule parameters and monitors system status. This work is covered by three patents (TW I769534B, TW I820977B and US 11979479B1).

SC26-NRE-127 PDF

The rapid rise of agentic AI systems, autonomous frameworks that bridge the gap between high-level reasoning and low-level execution, has begun to redefine modern cyberinfrastructure. Building on the DTN-as-a-Service (DTNaaS) lineage, including the foundational work “AIDTN: Towards a Real Time AI Optimized DTN System With NVMeoF,” the StarLight consortium introduces a multi-agent ecosystem that redefines data/process mobility. This architecture deploys distributed autonomous agents to plan, negotiate, and execute end-to-end missions across N × 400+ Gbps WAN infrastructures. Moving beyond static automation, these goal-directed entities employ adaptive reasoning and sophisticated policy to transform high-speed data/process transport into a self-orchestrating, intelligent cyberinfrastructure.

In this paradigm, large scale data/process transport emerges from collaborative agent behavior. Specialized agents continuously interpret telemetry, evaluate constraints, and dynamically reconfigure WAN, LAN, and cloud native resources, including Kubernetes orchestrated environments. Through advanced tool use, such as automated DTN provisioning, protocol selection, congestion avoidance, and real time anomaly mitigation, agents function as distributed operators capable of executing high performance data/process transport strategies at global scale. This shift lays the foundation for autonomous cyberinfrastructure capable of supporting next generation scientific instruments, real time global collaborations, and exascale data/process workflows.

The SC26 demonstration will showcase Agentic AIDTN through orchestrating data/process transport services across federated Kubernetes clusters and widely distributed sites. A Jupyter-based control plane allows scientists to collaborate with the Agentic AIDTN collective, auditing agent-generated plans, refining heuristics, and observing real-time self-coordination as the system optimizes Data Transfer Nodes (DTNs) for diverse scientific workflows.

Demonstrated capabilities include Memory-to-Memory and NVMe-over-Fabrics (NVMeoF) transfers using RoCEv2 and TCP across high-radix N × 400+ Gbps WANs. Building on the SC25 achievement of N x 360G+ Gbps aggregate throughput, this year’s work positions Agentic AIDTN as the primary engine for an autonomous global data-movement substrate, paving the way for resilient, self-directing infrastructure in the next era of data-intensive science.

SC26-NRE-128 PDF

GP4L (Global Platform For Lab) is a global overlay network composed of 100G/400G programmable switches and SmartNICs from the GNA-G AutoGOLE/SENSE Persistent Multi-Resource Testbed, NRP, FABRIC and the GEANT RARE initiative. It provides routing services using multiple open network operating systems, which implement traditional networking protocols and can be extended to integrate next generation features as an opt-in strategy. GP4L provides the means to implement production oriented feature experimentations and protocols, without impacting the rest of the network.

This showcase will include demonstration of advancements on the high performance software dataplane based on SmartNIC, VPP and multi control plane options. Demonstrate new integration capabilities to generate Digital Twin from GNA-G GREN Map topology files, and additionally support other NREs experimenting with at scale advanced/novel networking protocols and use cases, like PolKA (Polynomial Key-based Architecture) a stateless source routing protocol based on arithmetic operations over a polynomial encoded route label.

The Digital Twin use case consists of using network topology files, in GRENML data format from the GNA-G GREN Map initiative, to create digital twins with multiple profiles, from 1-to-1 copies of real networks to obfuscated versions for sharable virtual labs. The Digital Twins are created using the Netbox, Netreplica, Containerlab software stack.

The high performance software dataplane was presented at SC25 with focus on 400G performance evaluation and validation of multiple open/commercial control plane options. For SC26, the focus is to integrate SmartNIC QoS capabilities and validate VPP network emulation performance at 400G (delay, loss, jitter).

SC26-NRE-129 PDF

The research and education network at Sao Paulo (rednesp), formerly ANSP (Academic Network at Sao Paulo), connects dozens of research and education institutions in the State of Sao Paulo, Brazil also providing international connections to the USA and to Europe. After designing it in 2022, rednesp started deploying a new backbone, known as “Backbone SP” connecting major research and education institutions with 100 Gbps links: University of Sao Paulo (USP), University of Sao Paulo State (UNESP), University of Campinas (UNICAMP), Aeronautics Institute of Technology (ITA), Federal University of São Carlos (UFSCAR), Mackenzie University, Federal University of ABC (UFABC), Federal University at São Paulo (UNIFESP), National Institute for Space Research (INPE) and Data Processing Company of the State of São Paulo (PRODESP). All links are now fully operational including a 100 Gbps link to UNICAMP from the Brazilian Synchrotron Light Laboratory (LNLS) which is part of the Brazilian Center for Research in Energy and Materials (CNPEM), in Campinas (SP). Backbone SP is connected to the USA through two 100 Gbps and one 800 Gbps links (all shared with Amlight and RNP), recently updated from 2x200gbps, and to Europe through a 100 Gbps link using the Ellalink cable. The total bandwidth from Brazil to the US has risen to 1.2 Tbps in the beginning of 2026. There is also a connection to Chile (to the Vera Rubin Large Synoptic Survey Telescope (LSST)), and a connection to Buenos Aires in Argentina (RNP).

This year, the rednesp team will try to improve high-speed transmission efficiency. Last year, we achieved 81% efficiency on a 1.3 Tbps intercontinental link connecting São Paulo, Brazil, with the United States. While the available bandwidth will be lower this year, transmission efficiency is expected to improve by optimizing the transmission method and the number of streams, aiming for better utilization of the communication channel, which has high latency due to the great distances involved in intercontinental transmissions.

SC26-NRE-130 PDF

The Research and Education Network of Sao Paulo (rednesp), connects dozens of research and education institutions across the State of Sao Paulo, Brazil, while providing high-capacity international connectivity to the USA and Europe. Following its 2022 design phase, rednesp began deploying a new backbone infrastructure known as “BackboneSP”. This backbone connects ten major research and education institutions via 100 Gbps links, including: University of Sao Paulo (USP), University of Sao Paulo State (UNESP), University of Campinas (UNICAMP), Aeronautics Institute of Technology (ITA), Federal University of São Carlos (UFSCAR), Mackenzie University, Federal University of ABC (UFABC) and Federal University at São Paulo (UNIFESP), National Institute for Space Research (INPE) and Data Processing Company of the State of São Paulo (PRODESP). All links are fully operational, including a 100 Gbps connection between UNICAMP and the Brazilian Synchrotron Light Laboratory (LNLS), which is part of the Brazilian Center for Research in Energy and Materials (CNPEM) in Campinas. The backbone is connected to the USA through two 100 Gbps links and one 1.2 Tbps link, shared with Amlight and RNP.

The proposed experiment will evaluate high-throughput scientific file transfer using RDMA (RoCEv2) across a multi-domain, high-latency path between São Paulo, Brazil, and the SC26 show floor in Chicago. Building on promising local-lab tests between Data Transfer Nodes (DTNs), the team will measure end-to-end disk-to-disk performance over the intercontinental research and education network and compare it with conventional TCP-based transfers. The demonstration aims to characterize achievable performance, validate the infrastructure, and identify practical requirements for moving large scientific datasets across a wide-area network (WAN).

SC26-NRE-131 PDF

The SPHERE project proposes a live, multi-site demonstration showcasing SPHERE’s unique capabilities for cyber-physical systems (CPS), IoT experimentation, and bring-your-own-device/testbed (BYOD/BYOT) integration across distributed research infrastructure. The demo will highlight how SPHERE enables flexible, scalable, and interconnected experimentation spanning remote facilities, SCInet infrastructure, and portable edge deployments.

SC26-NRE-132 PDF

The American Science Cloud (AmSC) platform empowers AI-driven scientific advancements at scale by integrating data, models, and high-performance computing across DOE national laboratories. Through unified APIs, secure enclaves, and scalable services, AmSC fosters collaboration and innovation while seamlessly incorporating high performance computing centers, high performance networking, and commercial cloud solutions into a secure, adaptable hybrid environment. Serving as the backbone for the Transformational AI Models Consortium (ModCon), AmSC accelerates scientific discovery and strengthens U.S. leadership in science and technology. AmSC and ModCon the core infrastructure components of the overall Genesis Mission.

The DOE booth at SC26 will have a rotating set of Genesis demonstrations. Some will be technology focused and show AmSC capabilities, while others will be science focused and show how ModCon accelerates research. This NRE proposal is intended to support a variety of Genesis Mission demonstrations. Since the basis of AmSC is the integration of instruments, simulation, AI, data management, and networking, the SC26 NRE should support all these facets.

The DOE booth will have a small experiment or computer simulation generating data. The DOE booth will also have a display of AmSC resource status, so a low-latency link to all resources used in the demonstrations is needed.

Nodes in the SCinet cluster and nodes in the NRE hotel booth will be configured to run as an AmSC resource for storage and computing. The DOE booth, SCinet cluster and will need 100Gbps links to ESnet to enable workflows with resources at Argonne National Laboratory, Oak Ridge National Laboratory, Jefferson Laboratory and Lawrence Berkeley National Laboratory. In addition, connections to Amazon Web Services and Google Cloud are needed to reach AI resources.

Demonstrations will like include many of the AmSC infrastructure partners:

• VEE-ARIES: Virtual Emulation Environment for large energy projects
• DeepLynx: digital engineering integration hub
• EDX/MACH: platform for fossil energy R&D data and tools
• Fermi Data Platform: host and serve curated data sets
• HADIS: Hardware-accelerated AI-based Distributed Inverse Solver
• Femtomind: foundational infrastructure for QCD science
• S3DF: SLAC Shared Science Data Facility
• SCDF: Scientific Computing and Data Facilities
• Stellar-AI: combined HPC and AI for fusion energy research and deployment

SC26-NRE-133 PDF

Quantum communications have evolved from theoretical constructs to metropolitan-scale demonstrations. As the field matures, it is poised to transform quantum computing, much like how classical networks reshaped classical computing. However, scalability remains a critical challenge, particularly in terms of platform diversity, node count, communication distance, and application scope. The InterQnet project pursues a two-pronged strategy to advance the state of the art in scalable quantum communications. In InterQnet-Achieve, we focus on practical realization of a heterogeneous quantum network by building and then integrating a prototype first-generation quantum repeater with an error mitigation scheme and a centralized automated network control system. In InterQnet-Scale, we focus on a systems study of architectural choices for scalable quantum networks by developing futuristic models of quantum network devices, advanced error correction schemes, and entanglement protocols. In this demonstration, we will showcase our progress towards the InterQnet-Achieve goals through prototypical systems for remote entanglement distribution between heterogeneous qubit platforms deployed across the Argonne Quantum Network Testbed (ArQNet).

SC26-NRE-134 PDF

This demonstration runs a Quantum Key Distribution (QKD) link over production metro fiber between the Equinix facility at 350 E Cermak and the SC26 exhibit floor at McCormick Place, and uses the keys it produces to encrypt an optical wave on Internet2 transport equipment. A pair of ID Quantique Clavis XG systems exchanges keys over a dedicated dark-fiber strand. Those keys are delivered through a key management system to Internet2 Waveserver 5 encryption on a wave carried on Internet2 Ciena 6500 gear between 350 Cermak and the SCinet NOC. The chassis at both ends carry production traffic; the demo runs on its own module.

The point of the exercise is the quantum channel itself. QKD does not stop someone from tapping the fiber; it makes the tap visible. An IP-controlled attenuation device on the quantum strand in the SCinet Research Hotel lets us introduce loss on demand and show what the QKD system does when the channel is disturbed: the quantum bit error rate rises, the secure key rate falls, key sessions drop, and the encryptor is left without fresh keys. Visitors watch that happen on a display in the Internet2 booth.

SC26-NRE-135 PDF

perfSONAR is the de facto network measurement toolkit for research & education networks. Building on SC25-NRE-113 (perfSONAR for Community Access Broadband Initiatives), this NRE extends those methods to additional network operators across Cook County (SC26’s host region), with the goal of bridging the digital divide by connecting community-built perfSONAR nodes into the SC26 show floor.

Beyond the technical demonstration, this effort will engage and train Chicagoland participants in practical R&E networking skills through hands-on server-building workshops, preparing them for entry into registered apprenticeships or related cyberinfrastructure career pathways.

SC26-NRE-136 PDF

The SC26 SCinet Xnet team is building a White Rabbit subnanosecond time synchronization network between the SCinet NOC and SCinet Research Hotel. Xnet team’s focus will be on experimentation with perfSONAR one-way active measurement protocol traffic getting timestamped via FPGA. This NRE will allow technologists outside of the main Xnet effort to experiment with the SC26 White Rabbit infrastructure at SCinet Research Hotel. This “Precision Timing Plugfest” initiative will create an environment where SCinet’s multivendor integration spirit is alive & well, without adding additional demands to the Xnet project scope.

SC26-NRE-139 PDF

Data-Intensive Science (DIS) workflows, such as the Vera C. Rubin Observatory Long-Haul Network, move large datasets continuously over high-latency intercontinental links that cross several Research and Education (R&E) domains. These networks often offer multiple paths with different capacities and delays, yet traffic is typically confined to a single route or spread by Equal-Cost Multi-Path (ECMP) hashing, which ignores both path capacity and congestion and allows large flows to collide on the same link.

This demonstration presents PATH, a host-controlled approach for high-performance data transfers over a stateless network core. In PATH, the Data Transfer Node (DTN) itself selects the end-to-end path of each transfer among a set of pre-computed paths, encoded as RouteIDs, moving it to another path with no reconfiguration of transit nodes. Unlike hash-based host repathing, PATH lets the DTN select the complete forwarding path explicitly: each flow carries a precomputed RouteID, while transit nodes perform only stateless forwarding based on polynomial arithmetic over the binary Galois field GF(2).

PATH builds on PolKA (Polynomial Key-based Architecture), a source-routing scheme in which each core node is identified by a polynomial (nodeID) and each path is encoded as a single route label (RouteID). Every node obtains its output port as the remainder of dividing the RouteID by its own nodeID, so no per-flow or per-path state is kept in the core. At the DTN, programs written in eBPF (extended Berkeley Packet Filter) run in the Linux kernel data path (or in the NIC) and stamp each flow with the RouteID of the chosen path, so that path control stays with the host that owns the transfer.

SC26-NRE-140 PDF

KREONET (Korea Research Environment Open Network) operates as a backbone network with a minimum of 200 Gbps to a maximum of 4.2 Tbps per-link capacity, anchored by 17 regional network centers nationwide covering 200 R&E member institutions. NDeX is a dedicated infrastructure that interconnects supercomputers, AI data centers, and private/public cloud data centers at high speed, and is also directly interconnected with KREONET. Researchers can access data centers at high speed through KREONET-NDeX.

KREONET has developed a Software-Defined eXchange (SDX) technology to provide researchers with application-centric SDX automation technology through network awareness. Rather than relying on conventional BGP-based network path policy configuration, the NDeX-SDX platform collects network data including packet loss rate, per-port utilization, and BGP information (via SNMP, sFlow, and BGP) and upon forwarding this data to the Root Server, gains awareness of per-port conditions across the SDX network. Based on this awareness, the platform automatically generates optimal network path policies and triggers automatic network slicing, providing a flexible network operation environment through Application-Specific Peering technology.

The demonstration consists of two scenarios. In the first demo, the SDX network connected over a 100 Gbps link establishes a network connection with StarLight in the United States. Scientific data held by KREONET member institutions needs to be selectively routed to different destinations based on the nature of the data and its intended use. To address this, the demonstration showcases Application-Specific Peering technology, in which specific data is forwarded to the Daejeon DTN (Data Transfer Node) server, while other data is branched and transmitted to the U.S. DTN server.

The second demo utilizes KREONET and the NDeX network to connect NDeX with StarLight in the United States over a 400 Gbps link, demonstrating bandwidth utilization through data transfers between DTN servers in Korea and the U.S. The DTN node of the Korea Astronomy and Space Science Institute (KASI) in Daejeon, Korea, is connected to NDEX at 400 Gbps, enabling the transfer of LSST data between Korea and the United States.

Through both demonstrations, the NDeX-SDX platform implemented as a web interface provides each functional capability. The demonstrations validate the technology through the platform’s network control features, visualize real-time SDX network topology changes in response to control actions, and confirm data transfer performance over the 400 Gbps Korea-U.S. network.

SC26-NRE-141 PDF

Ciena’s WaveLogic 6 Nano (WL6n) is the latest offering in Ciena’s Coherent pluggable product line. Packaged in OSFP800 and QSFP-DD800 form factor with several transmission modes, the Ciena WL6n plug can establish error free channels over extended distances. WL6n enables a wide range of applications from 800ZR metro Data Center Interconnect (DCI) to 800G long-haul and 400G ultra long-haul. Based on 3nm CMOS and using Probabilistic Constellation Shaping (PCS), the latest WL6n variant introduces 135 GBaud modulation to achieve 800Gbps over distances greater than 2000km. For this demonstration at SC26, the WL6n gear would be deployed in the Ciena booth and connected to the Internet2 line system via SCinet. The channel would be looped in a Kansas City remote data center and eventually terminate in the Ciena booth on the showfloor. Ciena is expecting the signal to establish an 800Gbps channel over > 2000 km distance of terrestrial fiber via a remote loop. The line system will be EDFA-based only and will not contain any Raman amplification making the distance challenge that more difficult.

SC26-NRE-143 PDF

This demo showcases a Remote Embassy: a trusted network node deployed on infrastructure outside the home organization’s administrative control. The setup will consist of a K8S cluster distributed across two or more sites where one is named the home organization and manages the remote deployments through a trusted, quantum proof communication channel. The K8s cluster hosts applications and orchestrates virtual and physical network functions as service function chains, to support a variety of use cases where applications need to be distributed also outside the home perimeter but need to remain under the control of the home organization.

To protect communication between sites, we deploy the Fortaegis F1NE devices as physical network functions. Each device incorporates an intrinsic hardware security module (iHSM) that derives its identity from a physically unclonable function (PUF): a circuit that uses physical variations introduced during manufacturing to generate a device-specific identity. The devices authenticate each other and encrypt selected traffic exchanged between sites, at high speed.

Loom is a K8s-native orchestration framework for deploying and managing application workloads and network functions across heterogeneous software and hardware targets. It provides declarative abstractions for network functions, programmable software and hardware targets, and service function chains. In the demonstration, Loom orchestrates the network functions and configures a service chain that steers application traffic through them. Through a target-specific agent, F1NE’s encryption and decryption functions are integrated into the service deployment, protecting traffic exchanged between sites.

SC26-NRE-144 PDF

The success of AlphaFold demonstrated the importance of AI in sciences. However, the project could not have succeeded without the pdb, a large, open database of genomic and spatial structure of proteins. This database took almost 20 years and many billions of dollars to assemble. Google would never have considered to create such a dataset for the sake of performing AlphaFold.

This illustrates an important point: in the era of AI it is impossible for universities, academic institutions to compete with the hyperscalers, who invest hundreds of billions every year in hardware. For this reason, creating unique, high value, long lived data sets like the pdb remains a safe place for academics. Yet, universities do not have 20 years or 20 $B for this task, either.

The question is, how can we create a new, agile environment at our research institutions which can accelerate and support the creation of such data sets, but at a much lower cost.