A collaboration between the University of South Carolina, the Great Plains Network (GPN), and EPOC (a joint effort between Indiana University and ESnet) recently sponsored a two-day workshop on programmable data plane switches, with specific emphasis on Programming Protocol-Independent Packet Processors (P4).
Data plane programmability has attracted significant attention from the research community and industry due to its ability to enable programmers to run customized packet processing functions in the data plane, but so far there has been limited training available on P4.
P4 is a domain-specific language for network devices, specifying how data plane devices (switches, NICs, routers, filters, etc.) process packets. Using P4, application developers and network engineers can implement specific behavior in the network, enabling changes to be made in minutes instead of years.
“With FABRIC coming online and GPN as a host of one of the nodes, this P4 Workshop was a very welcome educational opportunity for our community” said James Deaton, the executive director of the Great Plains Network, one of the sponsors for the workshop.
More than 200 people attended the February event, which was presented via the University of South Carolina’s cybertraining system and offered free of charge. ESnet’s Jason Zurawski was among the organizers. Sessions covered the fundamentals of P4 programmable switches, FABRIC’s national cyberinfrastructure, use of P4 switches on campus networks. There were also hands-on sessions covering P4 building blocks, parser implementation in the data plane, populating match-action tables at runtime, and others.
While ESnet staff are known for building an ever-evolving network that’s super fast and super reliable, along with specialized tools to help researchers make effective use of the bandwidth, there is also a side of the organization where things are pushed, tested, broken and rebuilt: ESnet’s testbed.
For example, in conjunction with the rollout of its nationwide 100Gbps backbone network, the staff opened up a 100Gbps testbed in 2009 with Advanced Networking Initiative funding through the American Reinvestment and Recovery Act. This allowed scientists to test their ideas on a separate but equally fast network so if something crashed, ESnet traffic would continue to flow unimpeded across the network. Six years later, ESnet upped the ante and launched the 400Gbps network — the first science network to hit this speed — to help NERSC move its massive data archive from Oakland to Berkeley Lab.
Eric Pouyoul is the principal investigator for the testbed and the things he’s learned on past projects can be applied to others. His most recent project also pushed the boundaries of what the organization does in supporting DOE science. With funding from the lab’s Nuclear Physics Division, Pouyoul developed a pair of uniquely specialized data processing systems for the GRETA experiment, short for Gamma Ray Energy Tracking Array. The gamma ray detector will be installed at DOE’s Facility for Rare Isotope Beams (FRIB) located at Michigan State University in East Lansing.
When an early version of GRETA goes online at the end of 2023 it will house an array of 120 detectors that will produce up to 480,000 messages per second—totaling 4 gigabytes of data per second—and send them through a computing cluster for analysis. Not only did Pouyoul write the software for the first stage that will reduce the amount of data by an order of magnitude—in real-time—he also designed the physics simulation software to generate realistic data generation to test the system.
For the second data handling phase of GRETA, called the Global Event Builder, he wrote the software that will take all of the data from the first phase and, using the timestamps, aggregate them in order, as well as sort them by event. This data will then be stored for future analysis.
Even though he designed and built the systems to simulate the behavior of the nuclear physics that will occur inside the detector, “don’t expect me to understand it,” Pouyoul said. “I never did anything like this before.”
GRETA is the first of its kind in that it will track the positions of the scattering paths of the gamma rays using an algorithm specifically developed for the project. This capability will help scientists understand the structure of nuclei, which is not only important for understanding the synthesis of heavy elements in stellar environments, but also for applied-science topics in nuclear energy, nuclear forensics, and stockpile stewardship.
“This has been my most exciting project and it only could have happened here,” he said. “I think it takes me back to the origins of the Lab when scientists and engineers worked together to create new physics. We know it will work, but we don’t even know how the results will turn out, we don’t know what will be discovered.”
Before joining ESnet at Berkeley Lab 11 years ago, he had worked in the private sector. At one point in his career, he wrote code for control systems for nuclear power plants. Looking back, he estimates that maybe three lines of his code made it into the final library. He’s quick to point out that he doesn’t consider himself a software engineer, nor does he think of himself as a network engineer. At ESnet, those engineers are responsible for designing and deploying robust systems that keep the data moving in support of DOE’s research missions.
“I really like to work with prototypes, one-time projects like in the testbed,” he said. “I know how to build stuff.”
He developed that skill as a high school student in Paris, where he preferred to roam the sidewalks, looking for discarded electronics he could take home, repair, and sell. He did manage to attend classes often enough to pass his exams and graduate with a degree. That was the only diploma he’s ever received.
Since then, he’s learned by working on things, not sitting in lecture halls. Some of it he picked up working for a supercomputing startup company. He learned how to tune networks for maximum performance by tweaking data transfer nodes, the equipment that takes in data from experiments, observations, or computations and speeds them on their way to end-users.
He sees the GRETA project as a pilot and it’s already drawing interest from other researchers. The idea is that if ESnet can work with scientists from the start, it will be more efficient and effective than trying to tack on the networking components afterward. Pouyoul looking forward to the next one.
“I’m really not specialized, but I do understand different aspects of projects,” he said. “I only have fun when I’m not in my comfort zone — and I had a lot of fun working on GRETA.”
FABRIC, a project funded by the National Science Foundation, announces the formation of a Scientific Advisory Committee (SAC) tasked with facilitating collaboration and providing scientific and technological review for the project. FABRIC will create a unique national research infrastructure for testing novel architectures aimed at building an extensible, more secure Internet.
With leadership from the Renaissance Computing Institute (RENCI) at the University of North Carolina at Chapel Hill, the FABRIC project will build a large-scale platform with storage, computational and network hardware nodes across the country that are connected by dedicated high-speed optical links. FABRIC will also link major national research facilities such as universities, national labs and supercomputing centers that generate and process enormous scientific data sets.
The SAC will help guide the project by providing recommendations and critical feedback. Initially, the focus will be on reviewing the FABRIC design to ensure it can meet the diverse research needs of the future. The committee will also facilitate critical partnerships between collaborating institutions both within and outside of the US. As work progresses, the SAC will develop grand challenges that focus on solving key research problems using the FABRIC infrastructure.
“We are excited to have key research leaders across diverse career stages in fields such as networking, computing, software and security as our Scientific Advisory Committee,” said Inder Monga, co-PI of the FABRIC project and executive director of the U.S. Department of Energy’s Energy Sciences Network at Lawrence Berkeley National Laboratory. “The work is progressing well with FABRIC, and we look forward to the committee’s guidance on building an infrastructure that can facilitate testing of radical new ideas and approaches that will help lay the groundwork for the future Internet.”