How does testing affect the availability of aging software systems?

This paper proposes an approach to examining how testing affects the operational behavior of aging software systems. Such an approach requires models for the testing phase and the operational phase that explicitly account for crash failures due to both aging-related and non-aging-related bugs. We develop appropriate semi-Markov models and derive expressions for computing the respective transient and steady-state probabilities needed. Our numerical examples suggest that disregarding the effects of non-aging-related bugs can result in wrong conclusions about the testing phase and the operational phase. Moreover, we show how to combine the two models for a joint analysis in which metrics of interest concerning the operational phase, like the optimal rejuvenation rate, are random variables whose distributions are influenced by the potential outcomes of testing.