Feb. 20, 2024 — Lawrence Berkeley National Laboratory (Berkeley Lab) is “where it all began” for this year’s Society for Industrial and Applied Mathematics (SIAM) Activity Group on Supercomputing Early Career Prize recipient, computer scientist Giulia Guidi.

She first came to the Lab from Italy as an intern in 2017 while completing her master’s degree in biomedical engineering. Guidi liked it so much she decided to apply to UC Berkeley for her Ph.D. program, where her link to Berkeley Lab continued as a graduate research assistant in the Computational Research Division.

Advised by Aydın Buluç and Kathy Yelick, Guidi began her computer systems research with a focus on the challenges of large-scale computational biology. It continues to be her focus today and is the basis for her early career prize.

“My Ph.D. dissertation was on how to use sparse matrix-based computation to represent biological algorithms and biological computation, in particular on the de novo genome assembly, which was a topic of interest for the ExaBiome project at Berkeley Lab,” said Guidi. “My work has continued to focus on large-scale computational sciences, so I’ve always had my connection to Berkeley Lab.”

“It’s such an honor to receive this prize,” she added. “It’s a very big recognition. I’m just at the beginning of my career, so it is really exciting that the work I’ve been doing has been recognized by the community.”

Guidi is currently an affiliate faculty in Berkeley Lab’s Applied Mathematics & Computational Research (AMCR) Division and an assistant professor in Cornell University’s Department of Computer Science. SIAM’s Activity Group on Supercomputing awards the Early Career Prize biannually to an individual in their early career for outstanding research contributions in the field of algorithms research and development for parallel scientific and engineering computing.

“Computational biology is a key area of the rapid growth of computing,” said Guidi. “The growing volume of data and increasing complexity have outpaced the processing capacity of single-node machines in these areas, making massively parallel systems an indispensable tool.”

Novel Approaches to Performance and Scalability

But it’s not trivial to optimize the performance of biological algorithms, she added. The challenges of parallelism on high performance computing (HPC) systems require novel approaches to maintain optimal performance and scalability, which Guidi is approaching through the use of sparse matrices. She performs much of her research on the Perlmutter supercomputer at the National Energy Research Scientific Computing Center (NERSC) at Berkeley Lab.

“A lot of data in science and engineering is in the form of graphs, and graphs by definition are irregular structures; chemical structures and biological networks don’t have shapes that are easy to represent,” said Guidi. “When you have irregularity in your data structure, that makes computation harder to parallelize.” Parallelization is easier when you have the same amount of work to assign to each “worker” in the HPC system, she explained, “but when you have irregular structures it becomes much trickier to determine how much work each worker should do to maximize load balancing.”

Guidi describes the sparse matrix as how a graph is represented to a computer, giving it just a file with the idea of the nodes and a connection between the nodes. “We take something that’s irregular and we represent it to a computer as something that’s a bit more regular and allows us to exploit parallelization techniques for linear algebra or to parallelize more easily,” she said.

Making HPC More Accessible for Science

As Guidi continues her work with sparse matrix computation, she’s now looking at it from a core HPC perspective, with a broader set of biological applications in mind rather than a specific one. The goal, she said, is still the same—to make HPC more accessible, widely available to scientists across the board, and able to run on a local cluster, eliminating the need for time at a supercomputing facility.

“That’s always my ‘north star’ when I think about research—how to make it more accessible,” said Guidi. “Because it is becoming more and more clear that thinking of HPC as a luxury is no longer valid; I think we are at an inflection point where scientists are realizing that it’s something they need for most applications.”

Guidi develops systems and conducts much of her research at supercomputing facilities such as NERSC and emphasizes that it is important to ensure systems work on the leading supercomputing facilities available. But the goal is to make them eventually more accessible and available to a wider audience. “There are few applications that need thousands of nodes, but plenty that would do really well with 16-100 nodes, which is the typical size of an institutional HPC cluster,” said Guidi.

“Very practically speaking, it’s a matter of not only being able to get data processing results faster—optimizing HPC so that what takes a couple of days instead takes a few hours or less, for example—but also enabling scientists to do things that they currently don’t simply because they take too long computationally.”

“This prize makes me motivated to keep moving this research forward; it feels like a propulsion force,” said Guidi. “Really, it’s all I’m doing in my work—trying to contribute and make things better.”

About Berkeley Lab

Founded in 1931 on the belief that the biggest scientific challenges are best addressed by teams, Lawrence Berkeley National Laboratory and its scientists have been recognized with 16 Nobel Prizes. Today, Berkeley Lab researchers develop sustainable energy and environmental solutions, create useful new materials, advance the frontiers of computing, and probe the mysteries of life, matter, and the universe. Scientists from around the world rely on the Lab’s facilities for their own discovery science. Berkeley Lab is a multiprogram national laboratory, managed by the University of California for the U.S. Department of Energy’s Office of Science.

---

Source: Keri Troutman, Berkeley Lab