A Markov Regenerative Model of Software Rejuvenation Beyond the Enabling Restriction

Software rejuvenation is a proactive maintenance technique that counteracts software aging by restarting a system or some of its components. We present a non-Markovian model of software rejuvenation where the underlying stochastic process is a Markov Regenerative Process (MRGP) beyond the enabling restriction, i.e., beyond the restriction of having at most one general (GEN, i.e., non-exponential) timer enabled in each state. The use of multiple concurrent GEN timers allows more accurate fitting of duration distributions from observed statistics (e.g., mean and variance), as well as better model expressiveness, enabling the formulation of mixed rejuvenation strategies that combine time-triggered and event-triggered rejuvenation. We leverage the functions for regenerative analysis based on stochastic state classes of the ORIS tool (through its SIRIO library) to evaluate this class of models and to select the rejuvenation period achieving an optimal tradeoff between two steady-state metrics, availability and undetected failure probability. We also show that, when G EN timers are replaced by exponential timers with the same mean (to satisfy enabling restriction), transient and steady-state are affected, resulting in inaccurate rejuvenation policies.