Evaluating Reconfiguration Impact in Self-Adaptive Systems -- An Approach Based on Combinatorial Interaction Testing

Self-adaptive software adapts its behavior to the operational context via automatic run-time reconfiguration of software components. Particular reconfigurations may negatively affect the system Quality of Service (QoS), and therefore their impact over the system performance needs to be thoroughly evaluated. In this paper, we present an approach, based on Combinatorial Interaction Testing (CIT), that generates a sequence of configurations aimed at evaluating the extent to which reconfigurations affect the system QoS. Specifically, we transform a Classification Tree Models (CTM) of the configurations domain to a Constraint Satisfaction Problem (CSP) in ALLOY, whose solution is a sequence of reconfigurations achieving T-wise coverage between system features, and R-wise coverage between configurations in the sequence. The resolution of the CSP is performed by an incremental growth algorithm that divides the generation of the sequence into sub-problems, and merges the results into a final set of test configurations. Preliminary validation in a self adaptive vision system shows that our methodology effectively identifies critical reconfigurations exhibiting a high variation in QoS. This result encourages the use of CIT as a strategy to evaluate the performance of self-adaptive systems.