SCinet at Scale: Building the Network Infrastructure for Discovery

When people first hear about SCinet, they often think of cables, racks, switches, and blindingly fast connections. And that is true. SCinet is the world’s fastest temporary network, a live testbed for pushing global science forward. But behind every terabit-per-second link and every experiment running across continents, there is a community of people making it real.

SCinet is not just a showcase. It is a working laboratory, a teaching ground, and a collaboration engine. Each year, hundreds of volunteers come together: students, engineers, researchers, industry partners. Together they build something that does not exist anywhere else, a multi-terabit, fully programmable network that only lives for a few weeks yet pushes the boundaries of what is possible for science everywhere.

This year, the SCinet community is larger and more connected than ever. We are excited to be engaging at the local level through CommUNITY@SC25 and the TNTs programs, where we have just sent our invitations to this year’s successful applicants. These individuals will be paired with mentors across SCinet, learning not only how to configure and run a network at scale but also how to debug, problem-solve, and collaborate in real time, setting them up for future careers in this space. For many, this is their first taste of operating a global-scale infrastructure, and we are delighted to have them on board.

Working groups are also expanding SCinet’s scope. The newly formed Automation Team is deploying modern tools like Cisco NSO and Ansible, widely used in large-scale production networks. These tools will work to automate as much as possible, so engineers can focus on the things that matter most. The goal is to take these optimizations and lessons into not only future SC conferences but also into universities, labs, and research networks our volunteers contribute to.

The SCinet Cluster Program, building on its SC24 pilot, is taking a dramatic leap forward in 2025. Using the same power cap established last year, they are bringing 10 times the compute, 20 times the storage, and new AI-ready hardware. This setup will provide a shared resource powering experiments, tutorials, and pilots, including a chatbot test. Our expansion is possible thanks to partnerships with the National Research Platform, who are providing a bridge of resources before, during, and after the event.

And behind the scenes, dozens of teams are making the whole system work: logistics, power, physical security, routing, WAN transport (now delivering 13 Tbps to the show, up from 8.5 Tbps last year), and contributor relations, with over $66.5 million in loaned equipment and services from 38 industry partners.

The GNA-G Data Intensive Sciences Working Group is helping operationalize next-generation networks in support of data-intensive science. With SC25 on the horizon, the group has moved into full deployment mode: activating nodes, scaling data movement, and deploying smart networks with new operational paradigms across continents. 

This year, SCinet is directly supporting real scientific use cases and intercontinental demonstrations. High-speed transfer paths ranging from 400 Gbps to 2+ Tbps are being lit between Caltech (Los Angeles), UCSD (San Diego), FIU (Miami), Starlight (Chicago), CERN (Geneva), KAUST (Saudi Arabia), and multiple sites in Brazil and South Africa. These connections are supported by an evolving testbed of smart caching, programmable routers, and AI-assisted edge devices.

Caches at Caltech, UC San Diego (UCSD), and at many sites in the Open Science Grid hosted by Internet2 and ESnet are being upgraded for production workloads. The SENSE project is moving beyond proof-of-concept and into operational status for the Compact Muon Solenoid (CMS) and ATLAS experiments at the LHC, along with many other science programs. Ciena and Nokia optical platforms, and switches from Arista, Ciena, Cisco, and Dell are being reconfigured to support quality-of-service routing at scale. 

Smart edge devices and AI accelerators, from GPUs to smartNICs, DPUs, and FPGAs, are entering the stack. Future-facing designs from the National Research Platform are being developed in collaboration with many GNA-G groups across the U.S., Latin America, Europe, Asia, and Africa, aiming to push edge computation and global collaboration into new territory. 

Advanced intelligent network services are entering production, alongside novel approaches to traffic engineering. These include new forms of source-routing like PolKA, which uses polynomial functions to encode path information for efficient, scalable forwarding without relying on traditional IP routing tables. Complementary techniques for edge-based traffic control and flow balancing are being deployed across wide-area and intercontinental paths, enabling fine-grained optimization of network utilization. At the same time, machine learning approaches to network telemetry and operational decision-making are being tested, offering new levels of automation and adaptability across this emerging global fabric. 

The Starlight team is also deploying and testing 400 Gbps and 800 Gbps links with pluggable optics, including hybrid channels capable of supporting both quantum and classical signals.

Community building as secret sauce

SCinet is about preparing science for the data era. SCinet gives the community a place to test those capabilities at scale, with real workloads and real partners, before they are needed for the world’s most data-intensive experiments.

But perhaps the most important lesson is this: building networks is building community. Students learn from veterans. Vendors collaborate instead of competing. International teams debug side by side. And when the show floor goes live in November, the world sees not just blinking lights and bandwidth charts, but the results of people coming together to make discovery possible.

With less than three months to go, the SCinet team is still in full build mode. Network Research Exhibits (29 accepted so far) are lining up. Fiber, power, routing, and security are being finalized. Volunteers are being placed across teams. And testing, always testing, continues at every layer.

By the time SC25 opens its doors, the world’s fastest temporary network will be ready. And it will be more than cables and circuits. It will be a living example of what happens when a global community comes together to engineer the future of science.