摘要

As one of core technologies of software rejuvenation, analytical models provide a decision-making basis for implementing rejuvenation. This paper builds analytic models using stochastic reward nets with three different rejuvenation policies: non-rejuvenation, time-based rejuvenation, and time and load-based delay rejuvenation, and presents how system transits from one state into another. The results of numerical simulation experiments show: i) With the optimal rejuvenation interval for the virtual machine monitor (VMM) and the virtual machines (VMs) respectively, the latter two rejuvenation policies outperforms the non-rejuvenation one in terms of system availability. Software rejuvenation is an effective way to improve system availability. ii) Further, the time and load-based delay rejuvenation policy is better than the time-based one in terms of system availability and throughput. The system availability improvement obviously depends on the varying system loads.

全文