Family-based performance analysis of variant-rich software systems

Abstract

The analysis is meant for behavioral models of workflow-type software systems such as data centers or automation systems. We model these systems as a UML activity diagram with performance annotations to compute a performance prediction. A product-based (PB) analysis is not always viable, since we can have several variants, and each variant has its own performance model, which is why we cannot reuse the results of a single variant. Hence, we propose a family-based (FB) performance analysis relying on symbolic computation. The FB analysis can be significantly faster than PB analysis, especially for large-scale workflow model, thus enabling efficient calculation of large parameter spaces.