Software Rejuvenation Policy Modelling And Optimization Of The Virtualized Server

Separating the operating system and the applications from the computer hardware resources,virtualization has been used to increase resource utilization and ensure high availability of the system,thus decreasing operating costs in the data center.However,there exists the obvious drawback in the virtualized system architecture,that is7,the single point of failure.The feature of the structural dependence of the virtualized server has some impacts on its maintenance.On the one hand,the maintenance of the virtual machine manager affects the operation of all virtual machines on it;on the other hand,the maintenance behaviors of those virtual machines are competitive.These impacts bring new challenges to the maintenance decision management of the virtualized sever.Therefore,with consideration of the structural correlation between the multi-components of the virtualized server,the maintenance decision of the multi-components virtualized server and the maintenance decision of the multi-servers in data center have become the new problem to be solved.This dissertation focuses on computer software systems affected by software aging.Due to the inaccurate modeling of software degradation process,timeconsuming simulation modeling,and difficulty in solving high-dimensional models,this dissertation researches on modeling and optimizing rejuvenation policy for the virtualized servers and multi-servers in data center.The main research work is as follows:(1)Modeling and analyzing software aging process.A virtualized server based on Xen has been constructed.An experimental environment for detecting memory leaks and obtaining several performance indicators of system state has been implemented,using fault recovery policy,time-based rejuvenation policy and state-based rejuvenation policy respectively.Three kinds of system maintenance policies have been tested and experiment data have been collected.The maximum likelihood method has been used to estimate the parameters of the software aging process model.(2)Modeling and optimizing rejuvenation policy based on fixed-periodic detection and non-periodic detection under software measurable conditions.Under the condition that the degradation characteristics of the software system are given,the system degradation state has been obtained by fixed-periodic or nonperiodic detection.Based on all possible state transitions between maintenance decision points,the full probability of the steady state probability density function of the degradation state has been derived.The numerical solution to the probability density function has been given,and a long-term average availability model of the system based on the sermi-renewal process theory has been modeled.(3)Modeling and optimizing rejuvenation policy for multi-component systems with structural dependencies between VM and VMM in the virtualized server.The single component rejuvenation policy modeling method has been extended.On the basis of the analysis of all possible state transitions of the two components at the rejuvenation decision point,the system joint state probability density function has been derived,and the long-term average availability model of the system has been modeled.The optimization method has been presented,and the correctness of the model and the effectiveness of the optimization solution have been verified and analyzed by numerical experiments.(4)Modeling and optimizing rej uvenation policy for virtualized server deploying applications in redundant mode.Virtual machines quickly suspend,stop,restart,and migrate.On the basis of the above-mentioned features,the live-migration rejuvenation policy has been proposed.The probability that the application server can successfully migrate has been derived.The average availability and the average maintenance cost expression of the system have been obtained.And finally the rejuvenation decision has been modeled.Through numerical experiments,the correctness of the rejuvenation policy model and the adaptability of the optimization algorithm have been verified.(5)Modeling and optimizing dynamic rejuvenation policy for multiple virtual machines in cloud data center.The rej uvenation decision-making problem in multi-virtual machine with degradation failure and random failure is formalized into Markov dynamic programming(MDP).As for the time-consuming problem of ultra-high dimensional optimization,a heuristic rejuvenation policy based on health index has been proposed.The analytical expression of the health index function in the rejuvenation policy model has been derived.The rejuvenation policy has been evaluated through simulation experiments,and the correctness and effectiveness of the rejuvenation decision model have been verified.