DRUMS: Domain-specific Requirements Modeling for Scientists

Historically, scientists have developed software programs with limited resources and a strong focus on algorithms to generate scientific results and verify their approaches. With the rapid growth of hardware power, computer networks, and progress in software engineering, this brings a new challenge in scientific software development: how can scientists move from the development on the “algorithmic level” to the development of theories and methods that need the support of large-scale complex software systems. Many computational science and engineering (CSE) projects have been developing scientific software without a requirements specification. This is no longer possible in the development of large complex systems. Software engineers use requirements specifications to describe the complex systems being built, and bridge the gap of application domain (e.g. a scientific problem) and solution domain (e.g. a software implementation). The lack of requirements specifications hinder communication and collaboration between scientists and software engineers. To enhance the quality of scientific software and to adopt good software engineering practices, we claim that requirements specifications have to be used in CSE projects. However, formal requirements engineering methods are generally too heavyweight to be adopted in CSE projects, because scientists prefer to focus on producing scientific findings, not on spending time to learn the syntax for a requirements modeling language. On the other hand, it is even more difficult to ask a software engineer to develop requirements for scientific software, without obtaining deep domain knowledge. This dissertation describes DRUMS (Domain-specific ReqUirements Modeling for Scientists), which is a lightweight domain-specific requirements engineering framework. DRUMS provides abstractions that describe requirements in the scientific domain and tool support. Scientists can use these abstractions to effectively create and manage requirements without prior requirements engineering knowledge. In addition, DRUMS uses an automated approach that extracts requirements for scientific software systems to reduce the manual effort of requirements recovery and reuse. To demonstrate the applicability of DRUMS, experiments were conducted in three different scientific domains: seismology, building performance and computational fluid dynamics. The evaluation results show that DRUMS can effectively be used in early requirements engineering in these domains. Using DRUMS had a significant impact on the number of ideas generated in an early requirements engineering task, in comparison with a baseline practice. In addition, our requirements extraction approach outperforms naive Bayes classification on the same dataset.