Virtualized System Recovery Analysis Model And Its Application

As a proactive fault management technology, software rejuvenation rejuvenates a long-running system to its initial state or a relative healthy state in order to prevent the system from more serious faults in the future by suspending the system at the appropriate time and cleaning up its internal state.Thus software rejuvenation is an effective way to reduce system failures caused by software aging and to improve system availability.Software aging phenomenon appears in all software systems inevitably, including virtualized systems.It is necessary to study the rejuvenation policies suitable for virtualized systems.As an effective way to relieve the aging problem of virtual machine monitor,virtual machine live migration has a great influence on the rejuvenation process.This paper builds analytical models based on the virtual machine live migration, and analyzes the costs and profits of rejuvenation policies theoretically, which provides a theory basis for implementing rejuvenation.Our main works are as follows:1)We carefully study the existing virtual machine live migration algorithms and find that these algorithms can be very efficient in ideal envieonments and keep the balance of the downtime and total migration time.But it can not work well in an access-intensive environment or for large memory virtual machines.Therefore, we use a combinatory of page layering, active pushing, on-demand copying and compression, and propose an optimized algorithm called HLM (hybrid live migration). Experimental results show that even in the case of high dirty-page rate, HLM can work efficiently.2) To bulid three analytical models. For virtualized systems, we first build a time-based software rejuvenation model, a time and HLM-based rejuvenation model, a time and HLM and load-based rejuvenation model respectively. And then we studied the impact of varying parameters of models on the performance indicators (i.e.availability, etc.) of the system.3) Empirical study. We apply the three rejuvenation models into real systems for verification.