Leadership Computing Needs Leadership Software: Preparing for Exascale Success and Beyond

One legacy of the US Exascale Computing Project will be a collection of reusable scientific libraries and tools that provide portability across three distinct GPU-enabled computing systems based on AMD, Intel, and Nvidia accelerators. But we are just getting started with the need to support diverse platforms and scientific challenges.

Future leadership computing environments will include successors to systems such as Frontier, Aurora, and El Capitan, with increasing diversity in node architecture. Furthermore, these future systems will increasingly be part of a network of computing resources that include small-to-large-scale sensors and instruments, and a growing collection of large-scale data resources. The solution to next-generation scientific challenges requires sophisticated integration of all these resources. Productive and reliable software environments are key to success.

In this presentation, we first discuss current activities to produce the US Department of Energy scientific software stack for Exascale systems. Next, we describe emerging approaches to establishing a leadership scientific software organization. Finally, we discuss new activities underway to build a productive and sustainable scientific software community that centers on expanding the functionality, skills, and staffing strategies needed for long-term success in producing leadership scientific software, effectively and efficiently, for the coming decade and beyond.