Graduate student Ke Fan awarded prestigious fellowship in HPC

Ke Fan

PhD student Ke Fan was honored on August 14 by the Association for Computing Machinery and the IEEE Computer Society for her work in high-performance computing.

Fan was named one of two 2024 ACM-IEEE CS George Michael Memorial HPC Fellowship recipients. The George Michael Memorial Fellowship “honors exceptional PhD students throughout the world whose research focus is high-performance.”

Fan joined UIC’s computer science department in 2023. She and UIC Assistant Professor Sidharth Kumar, her advisor, were previously at the University of Alabama at Birmingham.

HPC aggregates computing power to achieve much higher performance than a typical desktop or laptop computer to solve complex problems more quickly and efficiently. Supercomputers are one of the most well-known types of HPC, consisting of thousands or millions of computational cores working in parallel.

“HPC applications are specifically designed to take advantage of the massive parallelism offered by these systems, requiring efficient inter-process communication to share and synchronize data, as well as advanced input/output (I/O) techniques for fast data handling,” Fan said. “The rapid advancement of computing technologies, particularly the emergence of Exascale supercomputers, has allowed for the exploration of increasingly complex and large-scale problems.”

Fan explained that this causes challenges related to the scalability of inter-process communication and parallel I/O, including efficiently utilizing parallelism at scale and addressing the gap between computational and I/O bandwidth.

Fan’s research addresses challenges in optimizing what is known as all-to-all data exchange, where each processor sends an individual message to every other processor by considering the characteristics of modern supercomputers. In response, Fan developed a class of hierarchical parameterized algorithms that are effective for both uniform and non-uniform communication workloads. These algorithms are general-purpose and can be applied across a wide range of HPC applications.

Fan has also made advancements in improving parallel I/O by tackling unbalanced workloads. The existing parallel I/O libraries typically leverage transformation techniques, such as compression, to mitigate I/O bottlenecks, which typically present load-balancing problems. These methods do not adequately address balancing problems, so Fan utilized spatially aware data aggregation techniques that can effectively facilitate load balancing.

Fan has also enhanced  the scalability of performance introspection frameworks, which profile HPC application performance, helping developers better understand data movement within HPC systems and identifying performance bottlenecks in parallel programs at scale.

“What’s impressive about Ke Fan’s work is that she has made algorithmic advancements within the complex domain of optimizing all-to-all communication—an area notorious for its scalability challenges in HPC,” Kumar said.

The Fellowship includes a $5,000 honorarium and travel expenses to attend the SC24 conference, where the Fellowships are formally presented.