Solomonik Recognized for Early Career Work Aimed at Improving HPC Performance
11/14/2018 10:51:57 AM
As a high school student, Edgar Solomonik was already focused on a career in computer science, working as an intern at a downtown Chicago trading firm.Commuting by train back and forth between home in the Chicago suburbs and the office in the Loop, he learned about the sort of dedication and thinking it takes to work and succeed.
“I already had a mindset for how to debug codes and kind of solve problems by myself, and how to work independently, the kind of work ethic that is associated with that,” said Solomonik, now an assistant professor at Illinois Computer Science and a graduate of the department (BS CS ’10).
Solomonik’s focus also has guided his research, which last month was recognized with the IEEE Computer Society’s TCHPC Award for Excellence for Early Career Researchers in High Performance Computing.
The award recognizes the 28-year-old Solomonik for research that “has developed more scalable numerical algorithms and high-performance computing libraries. His research group, the Laboratory for Parallel Numerical Algorithms, leads the development of the CYCLOPS library for tensor computations, which encapsulates optimized communication-avoiding algorithms using a high-level language for tensor algebra. CYCLOPS has enabled groundbreaking simulations of electronic structure and quantum circuits, while also being applied to graph and data analysis.”
Solomonik was nominated for the TCHPC Award by Professor James Demmel, the chair of Electrical Engineering and Computer Sciences at the University of California, Berkeley, and Solomonik’s PhD advisor.
In nominating his former student, Demmel noted the motivation of Solomonik’s work, and the advances for which he is responsible.
“Edgar’s work … is motivated by avoiding communication, i.e. minimizing data movement, since moving data (between levels of memory or over networks) is more expensive in time and energy than arithmetic,” Demmel wrote, noting that in some cases his work led to operations that improved efficiency by many times and that CYCLOPS is widely used in computational chemistry.
Solomonik noted that communication avoidance is a common theme in high performance computing.
“My research is especially focused on supercomputers and large clusters where you have many, many nodes connected across a large network and they’re sending a little bit of data from one machine to another, which can cost a few orders of magnitudes more than doing some on that machine,” he said. “So that kind of changes the perspective on how algorithms should be designed and also how software should be designed.”
Solomonik, as Demmel noted, created the first algorithms for fundamental matrix computations that achieve minimal communication cost.
The TCHPC Award is the second in just over a year recognizing Solomonik as a promising young scientist. He won the Householder Prize in 2017.
The latest honor, he said, both feels good and adds to the drive and focus that have so far characterized his career.
“I think adds drive, to try to do more and succeed more,” Solomonik said. “If people are recognizing what you’re doing is valuable, it makes you want to do more.”