The formal verification of finite-state probabilistic models supports the engineering of software with strict quality-of-service (QoS) requirements. However, its use in software design is currently a tedious process of manual multiobjective optimisation. Software designers must build and verify probabilistic models for numerous alternative architectures and instantiations of the system parameters. When successful, they end up with feasible but often suboptimal models. The EvoChecker search-based software engineering approach and tool introduced in our paper employ multiobjective optimisation genetic algorithms to automate this process and considerably improve its outcome. We evaluate EvoChecker for six variants of two software systems from the domains of dynamic power management and foreign exchange trading. These systems are characterised by different types of design parameters and QoS requirements, and their design spaces comprise between 2E+14 and 7.22E+86 relevant alternative designs. Our results provide strong evidence that EvoChecker significantly outperforms the current practice and yields actionable insights for software designers.
|Title of host publication
|30th IEEE/ACM International Conference on Automated Software Engineering (ASE 2015)
|Number of pages
|Published - 2015