Scaling size and parameter spaces in variability-aware software performance models

Abstract

Model-based software performance engineering often requires the analysis of many instances of a model to find optimizations or to do capacity planning. These performance predictions get increasingly more difficult with larger models due to state space explosion as well as large parameter spaces since each configuration has its own performance model and must be analyzed in isolation (product-based (PB) analysis). We propose an efficient family-based (FB) analysis using UML activity diagrams with performance annotations. The FB analysis enables us to analyze all configurations at once using symbolic computation. Previous work has already shown that a FB analysis is significant faster than its PB counterpart. This work is an extension of our previous research lifting several limitations.