Software rejuvenation of a fault-tolerant server subject to correlated failure

Society depends increasingly on software based services, necessitating that applications achieve a high level of reliability and availability to maintain customer satisfaction and preserve system safety. These service-oriented applications often operate continuously, which exposes the software to "aging" where resource deterioration such as memory leaks can lead resource exhaustion and ultimately system failure. To mitigate the impact of aging, software rejuvenation refreshes or restarts the system periodically to recover lost resources, thereby averting failures. Software rejuvenation may also employ fault tolerance to reduce the impact of individual failures on system availability. However, virtually all previous software rejuvenation studies assume that failures experienced by the individual servers comprising a fault tolerant system are statistically independent. This paper analyzes the impact of correlated failures on the optimality of rejuvenation schedules for a system composed of two parallel servers that can experience correlated failures. The examples illustrate the impact of correlation on the optimality of rejuvenation schedules to maximize availability and minimize down time cost. The results indicate that increasing the correlation between the failures of the servers increases the rejuvenation period when the objective is to maximize availability. However, the rejuvenation period decreases when the objective is to minimize down time cost. Thus, the proposed approach can objectively identify the optimal rejuvenation period that maximizes availability or minimizes down time cost.