Ana Veroneze Solórzano often jokes that she ended up in high-performance computing by accident. What’s not a joke is that she’s now transforming a field that itself is transforming the world.
High-performance computing, or HPC, doesn’t tackle small problems. Solving global warming and revealing the mystery of how galaxies are formed are par for the course with those who operate the world’s most powerful supercomputers. However, the field has problems of its own –– massive energy usage, gaps in privacy and inequitable access.
Solórzano, a computer engineering Ph.D. candidate at Northeastern University, is pushing HPC in bold new directions that could address those very challenges. The two prongs of her research, for which she recently received a George Michael Memorial HPC Fellowship, aim to make HPC more sustainable and private for the world’s supercomputer users –– and the millions of people that HPC research will potentially help.
“I saw [in HPC] an opportunity to learn more about the technical aspects of computing, on deeply how the architecture of a CPU or GPU worked but also [how it] caused an impact in the world,” Solórzano says.
Growing up in Porto Alegre, Brazil, a career in computing wasn’t part of Solórzano’s plan. Her self-deprecating joke about her winding path to HPC is partially true. She only ended up going into computer science because the coordinator for the international relations track she originally wanted to enter at a small private university in Brazil failed to show up for a meeting.
Instead, she had a run-in with the computer science program adviser, a happenstance meeting that changed the trajectory of her life.
Solórzano ultimately received her bachelor’s degree in computer science from the Federal University of Santa Maria. Solórzano later secured a position at the National Institute of Space Research in Brazil, working to apply an algorithm to identify particles in the universe. Her first taste of HPC only reinforced her need to further study advanced computing, and she eventually received her master’s degree from the Federal University of Rio Grande do Sul.
“They knew exactly the computational powers of the processors they were using and how to take advantage of that to cause an impact,” Solórzano says. “That’s when I saw if I get good at this specific field, I can do a lot of things.”
Now at Northeastern, Solórzano is tackling the field’s most pressing challenges at an institution that has gained a reputation for innovative and impactful computing research. She is the third Northeastern student and member of the Goodwill Computing Lab to receive the George Michael Memorial HPC Fellowship.
Solórzano’s work on one of Japan’s most powerful supercomputers, RIKEN’s Fugaku, is the first incentive-based energy control mechanism that has been put into production on an active supercomputer. The project is part of an ongoing collaboration between Northeastern and Japan’s RIKEN Center at Japan and Sandia National Lab in the U.S.
Supercomputers are open for researchers around the world to use for complex, boundary-pushing work that also uses massive amounts of energy.
Termed Fugaku points, Solórzano’s program allowed users to opt into three energy-saving methods while operating the supercomputer. In exchange, they could then spend “points” to access a priority queue that put them in front of the long line to use Fugaku.
The program ended up saving 10.7% of Fugaku’s energy in an initial round of testing and 13.26% in a second round as more people opted in.
“My group has been advocating for incentive-driven computer systems resource management for a long time, starting from early work in 2019 on coupon-based HPC storage bandwidth allocation,” says Devesh Tiwari, associate vice provost of research computing at Northeastern and Solórzano’s adviser. “I am very glad to see that this line of work is receiving acceptance and recognition.”
The goal, she says, is to create a bottom-up approach to making HPC more sustainable.
“From the supercomputer’s point of view, [energy usage] is a challenge, but until now the responsibility was all on the system administrators to make the decisions on how to manage that,” Solórzano says. “Then, for the first time, we worked on something that the users were also sharing the responsibility [for]. That’s what was the game changer.”
Solórzano’s other research addresses another lingering yet underdiscussed challenge around HPC: privacy.
Since supercomputers are shared by many users, privacy is a constant concern but one that has largely been underdiscussed. System administrators can scrub the name and data path for an application, but there is always a fingerprint, Solórzano says.
Her work on differential privacy, a method for protecting users’ data that involves adding “noise” to a dataset without losing its usefulness, aims to give people even more power to protect their data. With her tool, users can input their data and then essentially choose how much privacy they want. They then receive a privatized version of their data with a summary of how it has been transformed by the process.
A byproduct of this work, one that Solórzano is keenly aware of, is global data democratization. She hopes that by giving HPC administrators the tools to safeguard their data, they will be more willing to share it, especially with experts in countries like Brazil that have no supercomputers and rely on work done by those in the Global North.
“It’s something that is always in my mind,” Solórzano says. “Me and my adviser could have just written a paper on bringing differential privacy to HPC, but we don’t want just another paper. We want to open doors.”