�㽶��ý

A Universities Research Association (URA) Visiting Scholars Program grant is allowing �㽶��ý researchers to team with Fermilab scientists to develop software for a network data transfer switch that will allow scientists to compute data more efficiently.

Nik Sultana, assistant professor of computer science at �㽶��ý, received the grant to collaborate with Fermilab researchers who are working on the neutrino project build software that reads the data collected by switches in the system’s detectors in real time. Sultana, Research Software Engineer Nishanth Shyamkumar, Bjorn Ove Sagstad (M.S. CS 2nd Year), and researchers at Fermilab and CERN teamed up to write “,” which describes the programmable network prototype for the custom data packet format that is used by DUNE.

“This work is situated in an exciting problem space for finding creative solutions to the data networking problems of cutting-edge physics experiments,” Sultana says. “This space accommodates researchers with different expertise, ranging across physics, computing, and networking. It is particularly welcoming to students since the experts don’t have all the answers. We all work as students to navigate this problem space while learning new things along the way.”

DUNE’s detectors are designed to send very large data packets in a unique format. By programming the network equipment within DUNE’s network, the network itself could understand the data that it’s collecting rather than simply passing it into storage to be computed later.

This could lead to faster, smarter data handling for high-energy physics experiments.

“Ideas developed in this research can be adapted for other environments that involve transferring large quantities of latency-sensitive data streams,” Sultana says. “In the Chicago area, the fields that come to mind are electronic finance, online games, and supporting ongoing research on processing [artificial intelligence] workloads and on managing quantum computing workloads.”

The research was conducted on the testbed, an international infrastructure that enables cutting-edge experimentation and research for at-scale networking, cybersecurity, distributed computing, storage, virtual reality, 5G, machine learning, and science applications. The 29-site infrastructure is a distributed set of equipment at commercial spaces, national labs, and campuses. It includes large amounts of compute and storage, interconnected by high-speed and dedicated optical links. Its connections to specialized testbeds, the internet, and high-performance computing facilities create a rich environment for a wide variety of experimental activities.

“FABRIC served as a sort of virtual laboratory as we used it to prototype, test, and evaluate the research,” Sultana says. “In addition to hardware resources, FABRIC provides a flexible interface for testing our prototype and to evaluate it at realistic data rates. For our experiments on FABRIC, we were able to use actual detector data that was captured at Fermilab, which increased the fidelity of our research.”

Some of the challenges that the team faced included having to read data in mixed formats on switches that were not designed to handle these types of formats, reading very large data packets at high speeds, and using the features in these network switches for applications they were not designed for.

“This research is interdisciplinary, and it therefore requires us to learn things from outside our day-to-day research areas,” Sultana says. “This pushes you outside your comfort zone, and it can feel challenging, but it can also open paths through which research can have impact. In the case of this project, the �㽶��ý team learned a lot about how particle detectors work, how they produce data, and how that data is made available for analysis.”

Building on this prototype, the research could lead to future work such as designing custom hardware to process data or modifying how physics experiments format the data to make it easier to process in the network.