Piyush Sao, a research scientist in Oak Ridge National Laboratory’s Computing and Computational Sciences Directorate, has received the Best Paper Prize from the Society for Industrial and Applied Mathematics (SIAM) Activity Group on Supercomputing.
The prize recognizes the most outstanding paper on parallel scientific computing published in English by a peer-reviewed journal within the past four years. Sao and co-authors Sherry Li of Lawrence Berkeley National Laboratory and Rich Vuduc of Georgia Tech received the prize for their paper, “A communication-avoiding 3D algorithm for sparse LU factorization on heterogeneous systems,” published in 2019 by the Journal of Parallel Distributed Computing.
“It’s a truly humbling honor to receive this prize,” Sao said. “We continue to build upon the work discussed in this paper. The best paper prize will bring wider public attention to the paper, and I hope someone who might not otherwise look at our paper will do so and learn something new or unexpected.”
The paper proposes an algorithm to improve scalability of sparse lower-upper (LU) factorization on distributed-memory systems. Transferring data between nodes of a supercomputer – in this case, the Oak Ridge Leadership Computing Facility’s 27-petaflop Titan (now decommissioned) – quickly becomes prohibitively expensive for large simulations. The algorithm proposed in the paper cuts down on communication between nodes to avoid bottlenecks and speed up computation.
“Our best-case speedup was almost 27 times over the previous state of the art,” Sao said.
Sao will accept the prize and deliver a plenary lecture on the paper on February 25 at the SIAM Conference on Parallel Processing for Scientific Computing.
Sao earned his Ph.D. from Georgia Tech in 2018 and began his career at ORNL as a postdoc that same year. He now works in the Discrete Algorithms Group, led by Ramki Kannan. His work focuses on algorithms for supercomputing platforms with applications in machine learning and scientific computing.
Sao led the development effort for the DSNAPSHOT Project, an algorithm used to search medical literature for potential COVID-19 treatments that was also a finalist for the Gordon-Bell prize in 2020 and the DOE’s R&D 100 award in 2021. He also worked on the team that placed the OLCF’s flagship Summit supercomputer third on the Graph500 list, which ranks supercomputing performance for data-intensive applications.
“I’m grateful to the lab and all my colleagues for their support that makes this research possible,” Sao said.
UT-Battelle LLC manages Oak Ridge National Laboratory for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science. — Matt Lakin
