Palladio-based performance blame analysis

Performance is an important quality attribute for business information systems. When a tester has spotted a performance error, the error is passed to the software developers to fix it. However, in component-based software development the tester has to do blame analysis first, i. e. the tester has to decide, which party is responsible to fix the error. If the error is a design or deployment issue, it can be assigned to the software architect or the system deployer. If the error is specific to a component, it needs to be assigned to the corresponding component developer. An accurate blame analysis is important, because wrong assignments of errors will cause a loss of time and money. Our approach aims at doing blame analysis for performance errors by comparing performance metrics obtained in performance testing and performance prediction. We use performance prediction values as expected values for individual components. For performance prediction we use the Palladio approach. By this means, our approach evaluates each component’s performance in a certain test case. If the component performs poorly, its component developer needs to fix the component or the architect replaces the component with a faster one. If no omponent performs poorly, we can deduce that there is a design or deployment issue and the architecture needs to be changed. In this paper, we present an exemplary blame analysis based on a web shop system. The example shows the feasibility of our approach.