2018 Argonne Leadership Computing Facility (ALCF) Science Report

Abstract

This year sees the ALCF looking forward. As we prepare to enter the exascale era, the future of leadership computing lies in simulation, data, and learning. To best support this paradigm, we are striving to direct our efforts to develop the facility so that its growth synergizes with our users' research.In 2018, we expanded the Aurora Early Science Program (ESP), adding data analysis and machine learning projects to the simulation-based projects already underway.Collectively, these projects cover various combinations of techniques and approaches across a wide range of disciplines and goals. Common to all of them, though, is a substantial data challenge: at this point in computational science, data problems are no longer unique to projects that explicitly identify as such.The ALCF Data Science Program (ADSP) is yet another way we support projects that rely on advanced computational methods to enable data-driven discoveries. In addition, this year marked the first time that the INCITE program has explicitly sought data-and learning-based projects as part of its annual call for proposals.Partnering with the research teams supported by these programs allows us to take a collaborative approach to exploring not simply how to elevate the roles of data in computational science, but how to strike the appropriate balance between them and construct an architecture readily adaptable to each project's needs. A L c fWith a peak performance of more than 11 petaflops, the ALCF's Theta system is among the fastest supercomputers in the world for open scientific research.