Domain-Specific Languages for Ocean Modeling and Simulation

点击翻译按钮获取中文摘要

The physical, chemical and biological processes in the ocean significantly influence and contribute to the global carbon cycle, whose thorough understanding is crucial for future predictions of climate and climate change. As a consequence, ocean biogeochemical modelling and simulation is an important part of numerical climate research. With OceanDSL, we target this domain for research software engineers who typically have a background in physics, meteorology, chemistry, biology and oceanography.Computational science is a multidisciplinary field lying at the intersection of mathematics and statistics, computer science, and core disciplines of science. Computational science involves the development of models and simulations to understand natural systems answering questions that neither theory nor experiment alone are equipped to answer. Despite the increasing importance of so-called in-silico experiments to the scientific discovery process, well-established software engineering practices are rarely adopted in computational science. However, meanwhile the computational science community starts to appreciate that software engineering is central to any effort to increase computational science’s software productivity. For computational science, new domain-specific software engineering methods and techniques are required.A method that software engineering can offer to computational science is model-driven software engineering, which is a well-established approach for software engineering in the domains of information systems and embedded systems. Essential for model-driven software engineering are Domain-Specific Languages (DSLs). However, in high-performance computing, and especially ocean system modelling, DSLs are used seldomly, so far. This is caused by the lack of proper DSLs in that domain and the applicability of existing tooling in conjunction with the different roles and tasks involved in ocean system modelling.The objective of the OceanDSL project is to build and evaluate a DSL infrastructure for marine biogeochemical modelling and simulation, the configuration of simulation runs and the deployment on high-performance computing hardware. Core of the OceanDSL project is the development and comparative evaluation of these DSLs including their code generators (for external DSLs) resp. program libraries (for internal DSLs).Our research approach is twofold, combining constructive and empirical methods. With constructive research, we design both internal and external DSLs. For internal DSLs, we implement appropriate libraries. For external DSLs, we implement related infrastructure for editing and code generation. With empirical research, we plan a systematic empirical evaluation of the DSLs and its infrastructure together with ocean system modellers and research software engineers. A specific research question addresses the comparison of internal vs. external DSLs.