Towards a Catalog of Performance Smells for Parallel

Parallel computing architectures have been proven to significantly boost performance and are, therefore, being increasingly adopted in industrial applications. However, our experience with applications built on parallel computing architectures reveals that often they are not designed properly to leverage the power of these architectures and consequently suffer from performance ''smells''. We believe that an awareness of such smells can help practitioners design better and faster applications. However, our study of the existing literature on performance smells reveals the lack of a generalized comprehensive smell catalog that focuses on parallel computing. To address this gap, we have aggregated and cataloged commonly-occurring performance smells related to parallel computing. In this paper, we describe these smells along with their associated mitigation techniques, and attempt to draw out the relationships between different smells.