A Compositional Approach to Coordinated Software Rejuvenation of Component-Based Systems

In component-based software systems, micro-rejuvenation of individual components can be performed to limit the number of more time-consuming system macro-rejuvenations, requiring appropriate selection of rejuvenation times to actually reduce the system unavailability. We present a novel coordinated approach to micro-rejuvenation of software components, aimed at minimizing the system cumulative unavailability over time.Specifically, each component is periodically rejuvenated, and, if it fails, rejuvenation is scheduled after repair. To limit concurrent component rejuvenations, an efficient calculus is defined to derive the optimal rejuvenation offset of each component, based on the condition of system unavailability expressed as a static fault tree with AND and OR gates. An efficient compositional approach is also provided to derive the system unavailability, by leveraging numerical analysis of a stochastic model with underlying Markov regenerative process. The solution accuracy is exploited to analyze how components loose synchronization over time due to random rejuvenation and repair times, enabling the definition of a macro-rejuvenation policy. Experiments performed on several randomly and manually generated models show feasibility and effectiveness of the approach, enabling further extensions.