OLCF Brings Petascale Computing to 2017 APS March Meeting
DOE staff showcase scientific achievements in petascale computing at largest gathering of physicists
For staff at the Oak Ridge Leadership Computing Facility (OLCF), attending computing conferences and meetings is valuable, but sharing at scientific meetings where computing resources can have a direct impact on research challenges can be even more exciting.
For the fourth year in a row, the OLCF—a US Department of Energy (DOE) Office of Science User Facility located at DOE’s Oak Ridge National Laboratory (ORNL)—played a major role in the annual American Physical Society (APS) March Meeting, the largest gathering of physicists in the world. The March 13–17 event in New Orleans, Louisiana, afforded more than 10,000 scientists opportunities to present research, network, and share valuable scientific insights.
With the goal of bringing high-performance computing talks to a conference bursting with topics related to material and chemical physics, National Center for Computational Sciences Director of Science Jack Wells co-organized a focus session called “Computational Physics at the Petascale and Beyond.” The session showcased scientific projects and achievements in the realm of petascale science—those which are carried out by a system capable of executing one petaflop, or 1 quadrillion calculations per second. Session chairs were Wells, Nichols Romero from the Argonne Leadership Computing Facility, and Jack Deslippe from the National Energy Research Scientific Computing Center.
Any members of the society can present their work at the APS meeting through short talks, called contributed talks, or through contributed posters. Focus sessions, which have a topical focus, are proposed by APS members and selected by APS before the meeting as a way of zeroing in on new scientific discoveries, tools, and accomplishments in physics.
The March Meeting featured 50 parallel sessions with more than 8,100 papers presented. The meeting included 105 invited talks by leading figures in physics subdomains such as biological, chemical, materials, and others. For their focus session, Wells and his co-organizers recruited 36 contributed talks from around the world and recommended to March Meeting program leadership 3 invited talks. The focus session comprised 9 total hours of content.
“With our session, we really wanted to encourage people to present their work on petascale computing systems of today, and thoughts on the future—toward exascale—were also of interest,” Wells said.
The three invited talks featured Nicola Marzari, chair of theory and simulation of materials at the Ecole Polytechnique Federale de Lausanne in Switzerland; Cecilia Clementi, professor of chemistry and senior investigator of the Center for Theoretical Biological Physics at Rice University and codirector of the Molecular Sciences Software Institute (funded by the National Science Foundation); and Lin Lin, assistant professor in the department of mathematics at the University of California, Berkeley.
Marzari shared about quantum-simulation engines and the challenges of implementing these on large-scale, hybrid (CPU–GPU) parallel architectures such as Titan, the OLCF’s Cray XK7 supercomputer capable of 27 petaflops. Lin presented a numerical method for scaling density functional theory calculations for large-scale applications.
Clementi, a new OLCF user who was awarded computing time on Titan through an Innovative and Novel Computational Impact on Theory and Experiment allocation, is a physicist exploring macromolecular dynamics through a computational lens. In her talk, she presented approaches for “adaptive sampling,” which can be used to significantly speed up the simulations of slow dynamical processes in biomolecular systems such as protein complexes. Exploring these involved processes is crucial for understanding biological functions.
Because adaptive sampling approaches include the atomistic details of large biomolecules within complex environments, the simulation needed a large parallel computer such as Titan. These methods will ultimately extend the timescales for Clementi’s simulations of the immune system and further her research on the protein interactions responsible for autoimmune defects.
“Using Titan, we now have several thousand GPUs that are being used at one time,” Clementi said. “The communication between GPUs and CPUs takes time, so different computational strategies are required for different projects to ensure optimal distribution of these resources.”
Clementi said APS is important because it helps the physics community recognize that computation can benefit scientific domains other than computer science. “Many theoretical physicists may not realize the importance of computation,” Clementi said. “But software and simulation are important tools in modern science, and we can’t do our work without them. This is why bringing topics like petascale computing to APS is important.”
Wells—whose background is in physics—said the APS March Meeting is important for ORNL’s visibility, and he enjoys sharing the OLCF’s work with one of his scientific communities each year.
“Our supercomputers are important instruments of scientific discovery, so it’s crucial that the work done at the OLCF is presented in a visible way at the most popular scientific meetings in each community,” Wells said. “The APS March Meeting lets us connect with an invaluable community—that of materials physicists, chemical physicists, and biophysicists.”
Oak Ridge National Laboratory is supported by the US Department of Energy’s Office of Science. The single largest supporter of basic research in the physical sciences in the United States, the Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov.