Building a Consistent Taxonomy for Parallel Programming Models

Abstract

Parallel programming has been a challenge for developers and software engineers for over two decades now. To lower the complexity of parallel programs, a lot of different parallel programming models (like ACTORs) or supporting libraries (like MPI) have been introduced. These models and libraries support different features and have individual hardware requirements. As part of a software development process a software engineer has to choose which programming model or library is used and fits best for a specific use case. Usually this discussion is made in an early design phase and depends on multiple factors. To enable the software engineer to make a well-informed decision a taxonomy is required, that contains all necessary information. To find such a taxonomy we performed a Systematic Literature Review, within we found five taxonomies. However, the found taxonomies are inconsistent regarding structure, terms and included models. This paper discusses the five found taxonomies and we will propose a new taxonomy, that overcomes their shortages and combines their features. With our proposed taxonomy software engineers are able to make well-informed decisions already in early phases of the software development process.