PluginPlay: Enabling exascale scientific software one module at a time

Area of Science:

• Computational chemistry
• High-performance computing
• Software engineering for scientific applications

Background:

• Exascale computing presents significant challenges for scientific software, often necessitating extensive rewrites.
• Scaling to exascale demands more planning and optimization than petascale, requiring substantial resources and time.
• Sustainable computational science necessitates modular software, but leveraging it effectively remains challenging.

Purpose of the Study:

• Introduce PluginPlay, an inversion-of-control framework.
• Facilitate the development, maintenance, and sustainability of modular scientific software packages.
• Focus on design aspects influencing the performance of scientific software.

Main Methods:

• Development of an inversion-of-control framework named PluginPlay.
• Focus on design principles for modularity and performance.
• Application within the NWChemEx package as a case study.

Main Results:

• PluginPlay provides a sustainable approach to modular software development.
• The framework's design directly impacts the performance of the resulting scientific package.
• Demonstrates a method for managing complexity in exascale software.

Conclusions:

• PluginPlay is a generally applicable tool for computational science packages aiming for exascale.
• The framework supports sustainable development and maintenance of modular scientific software.
• Facilitates the transition of computational science packages to exascale computing environments.