Rheos A Domain-specific language for high-level sampling tasks in high-performance computing

Abstract

Many computations running on high-performing systems do not make use of the performance available. To solve this problem, software written to achieve strong scaling is needed. Copernicus is a system for execution of large-scale sampling tasks in high-performance environments. It aims to achieve strong scaling, regardless of underlying architecture. The system was originally developed to run large scale bio-molecular simulations. However, lacking an intuitive way of describing computational projects, the developers felt a need for an user-friendly text-based input for Copernicus. This master’s thesis describes a design and implementation of a domain-specific language to meet the need of a suitable input description for Copernicus. The language design is simple yet manages to capture the abstract model of how a computational project is executed. The language is strongly typed and inspired by elements from both functional programming and data-flow languages, making Rheos a powerful descriptive domain-specific language.