Abstract: The increasing use of computational systems has highlighted concerns about attributes that may influence the quality of service, such as performance, availability, reliability, and maintenance capacity. Failures in the software development process may impact these attributes. Flawed code and overall software misdesign may cause internal errors, leading to system malfunction. Some errors might be identified and fixed during the software testing process. However, other errors may manifest only during the production stage. This is the case of the software aging phenomenon, which is related to the progressive degradation that a software performance or reliability suffers during its operational life. This paper proposes a methodology for software maintenance that is tailored to identify, correct, and mitigate the software aging effects. If the source code can be modified and a new version deployed with minimal impact, thus data from aging detection is used for corrective maintenance, i.e., for fixing the bug the causes the aging effects. If the software cannot be fixed nor its version updated without long system interruption or other bad consequences, then our approach can mitigate the aging effects, in a preventive maintenance to avoid service outages. The proposed methodology is validated through both Stochastic Petri Net (SPN) models and experiments in a controlled environment. The model evaluation considering a hybrid maintenance routine (preventive and corrective) yielded an availability of 99.82%, representing an annual downtime of 15.9 hours. By contrast, the baseline scenario containing only reactive maintenance (i.e., repairing only after failure) had more than 1342 hours of annual downtime - 80 times higher than the proposed approach.
Details in this link.