Assistant Computer Scientist Min Si has been working in Argonne National Laboratory’s Mathematics and Computer Science (MCS) division conducting research in dynamic execution runtime and high-performance computing implementations. But her association with MCS dates back farther – beginning with her work at Argonne as a research aide in 2013 and subsequently as a guest graduate student while pursuing her doctoral studies.
After graduating from the University of Tokyo in 2016 with a Ph.D. in computer science, Min joined MCS as an Enrico Fermi Postdoctoral Scholar. This named Argonne postdoctoral fellowship is awarded to outstanding international scientists and engineers who are at early points in their promising careers. It is named after Enrico Fermi, for his involvement in and supervision of the experiment at Chicago Pile-1 that created the world’s first sustaining nuclear chain-reaction, which ultimately led to the founding of Argonne.
In the Q&A section below, Min Si discusses her interest in computer science, her ongoing research efforts, and her experiences at Argonne.
- What sparked your interest in computer science?
My mother is a software developer and a computer science teacher, so I was introduced to the field at an early age. I wrote my first toy program in middle school. While my undergraduate studies were in Japanese language and culture, I worked for a software company as a software developer after I graduated. Programming was a creative job for me at that time; however, I wanted to solve more challenging problems in advanced high-performance computing (HPC) systems. Thus, I decided to go back to school to pursue my Ph.D. in computer science.
- Why did you apply for the Enrico Fermi Fellowship, and what drew you to Argonne specifically?
My professor from the University of Tokyo introduced me to my supervisor at Argonne. It started as an internship as a research aide; I really liked my topic and the work environment at the lab. I ended up working on most of my Ph.D. work at Argonne as a long-term graduate student. I knew about the fellowship because of my affiliation with Argonne, and I decided to apply prior to my doctoral graduation.
- Can you tell us about your current research projects?
My main research topic is concerned with data movement. This broadly refers to moving data between devices, such as a CPU core and memory, or between a local computer and a remote machine. It proves to be a critical problem on a larger scale, specifically in supercomputers, because moving data between machines takes up a lot of time. My research focuses on minimizing the cost of data movement; the faster you can handle data movement, the shorter time it’ll take to complete computational work, thus saving time and energy.
I’ve been investigating irregular application and dynamic communication. Previously, HPC applications based on the Message Passing Interface (MPI) were usually written in a symmetric and regular structure. This means that at a given point of time, data would be exchanged or gathered among a known group of processing units. This model is inefficient for some dynamic applications where the pattern of data movement is not known from the start; that’s where supporting dynamic communication plays a major role. My work addresses the question of how to transparently enable fast data movement for dynamic applications by leveraging knowledge from applications and underlying architecture.
My work in data movement with regard to dynamic applications resulted in “Beehive,” my postdoctoral project. Beehive decomposes a big problem into small blocks, where each processing unit does one block of work. It serves as a dynamic execution environment for supporting irregular MPI applications. We know each processing unit’s role, and we can dynamically manage the scheduling of computational tasks over different physical CPU cores. This hides data movement and migrates the small tasks into different CPU cores to balance computation. “Casper” is one of the deliverable software results from Beehive; it can dynamically optimize the progress of data movement and is portable and easy to install. It significantly reduces the computing time of the widely used quantum chemistry application NWChem on several supercomputing systems. “Process-in-Process,” the other software component of Beehive, allows applications to efficiently utilize the shared memory resource on modern multi- and many-core systems. This work was recognized with the highly competitive Best Paper Award at the 27th International Symposium on High-Performance Parallel and Distributed Computing.
- Tell us a little bit about your work environment, specifically the MCS division, and how it’s conducive to your research projects. How does your position fit into your overall career goals?
I believe that the lab atmosphere at Argonne is ideal for producing deliverable software while also working on research. The flexibility between researching and producing deliverables is unlike industry and academia; it delivers the ideal balance between innovation and practical impact. I want to focus on software delivery and design new software that has an impact in the real world. Being a part of MCS is especially great for my research because we have easy access to the supercomputer and we can easily communicate with various system experts as well as computational scientists.
- What’s the best part about working in the MCS division?
It’s easy to collaborate with experts in different fields! We get the opportunity to meet people who are experts in their fields, and we’re given the chance to participate in large projects that contribute to advanced system infrastructure for flagship supercomputers.
- And lastly, what are your interests outside of computer science?
Apart from computer science, I like to learn about Asian transitional culture, languages and history. My hobbies include gardening, traveling and playing with cats!
For more information, visit Min Si’s web page: http://www.mcs.anl.gov/~minsi/