George Biros' research group pushes computational efficiency by programming with hardware in mind

ICES Professor George Biros stands with his code

For decades, scientists have been increasing the speed and efficiency of high performance computing (HPC) by cramming more and more transistors onto processing cores, a phenomenon described by the famous Moore’s Law.

However, there’s a limit to how small transistors can get. And Intel, one of the leaders in transistor miniaturization, gives it five years until the end of Moore’s Law. Without a way to increase computing power, high performance data analysis and simulation—from medical research to climate modeling —could plateau, limiting the complexity and the types of problems that we are able to solve.

Luckily, ICES researchers are developing new methods that could continue to boost computing efficiency. Instead of the brute force of transistors, researchers in the ICES Parallel Algorithms for Data Analysis and Simulation group are focusing on how to better direct high performance computing by designing new algorithms that utilize existing and near-future HPC technologies. In a way, it’s getting high performance computers to work smarter rather than harder.

“We really have to think about how to use the hardware algorithmically to go to the next level,” said George Biros, director of the group and professor of mechanical engineering.

The National Science Foundation is supporting the center's research into how the complex hardware of high performance computers can be better utilized to make programs run more efficiently. Dhairya Malhotra, a Ph.D. student in his final semester at ICES, has been an integral part of Biros’ team developing such novel methods.

Malhotra first started working with Biros in 2009 as an undergraduate summer intern at the Georgia Institute of Technology, where Biros was then a professor. In 2010, he was part of the research team led by Biros that won the Gordon Bell Prize—one of parallel computing’s highest honors—for developing a program for the numerical simulation of blood flow that adapted to efficiently run on different high performance computing systems. And in 2015, he won the George Michael Memorial HPC Fellowship, a prize that recognizes exceptional Ph.D. students in high performance computing, and is awarded jointly by the Association for Computing Machinery and the Institute of Electrical and Electronics Engineers.

“He’s a true scholar and he wants to do things right,” Biros said.

As a graduate student at ICES, Malhotra has continued developing algorithms for modeling complex fluids. These kinds of algorithms are essential to improving modeling of the flow of fluids with microscopic particles, from blood pulsing through the body, to oil and gas flowing through fractures in rock. A group led by Michael Shelley, a mathematics professor at New York University, is even using codes developed by Malhotra to study how cytoplasm flows between a parent and its daughter cells during cell division.

Specifically, Malhotra has helped develop algorithms for complex fluid flow governed by integral equations rather than differential equations. Biros describes integral equations as “specialized surgical tools” used to develop high-precision algorithms. The narrow focus, and high cost of running such algorithms has generally limited their development and use. But by leveraging methods that prioritize computational efficiency, Malhotra has been able to make integral equations an option for studying complex fluid flow phenomena.

“Working on challenging problems motivates me,” said Malhotra, who will be continuing his work as a post-doctoral researcher at New York University’s Courant Institute of Mathematical Sciences in the fall.

Still, while integral equations allow for more realistic depictions of specific phenomena, the flexibility and speedier processing of differential equations are valuable attributes in their own right. A long-term research project of Biros’ research group has been developing predictive algorithms for brain tumor growth. A new collaboration with Mateo Ziu, a neurosurgeon at the Dell Medical School, is allowing access to data that will help refine these algorithms further.

Whether it’s through differential or integral methods, the work in Biros’ group is improving the computational processes that are becoming more important as the age of Moore’s Law comes to a close, and playing an influential role in training the next generation of researchers, such as Malhotra, who will help push computational efficiency to new places.


George Biros receives grant to lead research on large scale algorithms for uncertainty quantification

The Biros Code for Efficient Programming

George Biros Bio

Posted: Aug. 8, 2017