UNM Assistant Professor of Computer Science Amanda Bienz has received an NSF CAREER Award to explore ways to enhance the performance of parallel applications on emerging supercomputers.

With the award, Bienz hopes to find the best ways to solve problems using supercomputers whose software capabilities need to evolve as quickly as the powerful hardware that is used by researchers and scientists across disciplines.

“The work is particularly important at this time because of the diversity of emerging parallel computers,” Bienz said. “Originally, algorithms were being optimized for single compute cores connected by a network interconnect.  Today’s computers consist of nodes, which each contain a large number of cores, some contain GPUs that do all of the computation,” she said. 

Such performance upgrades mean faster sharing of information, Bienz said.  These performance optimizations can be useful in a large variety of applications, she said. Her research, however, is particularly focused on reducing the cost of complex patterns of data movements, such as instances in which large volumes of data is communicated. It can also be applied in unstructured communication where each process sends data to a random subset of other processes. This has the potential to impact parallel applications that rely on numerical methods as well as neural networks,” she said.

While these performance optimizations can be useful in a large variety of applications, the research is particularly focused on improving numerical solvers such as algebraic multigrid and fast fourier transforms.  The research will also look at improving allreduce operations on heterogeneous computers in an effort to improve the efficiency of distributed neural networks.

“There is so much heterogeneity in the computers that it is very difficult to move data efficiently, which causes applications to run slow.  So, while computers are increasingly powerful, applications are often unable to take advantage of the emerging hardware,” Bienz said.

The $557,000 award, which is the most prestigious from the NSF for junior faculty, is for five years. It funds the work of an undergraduate and graduate student at UNM.

“My hope is that this project leads to methods for parallel applications to more efficiently utilize emerging architectures so that we can solve exciting new problems through computer simulations,” Bienz said.