Research On Self-optimization Method Of System Depen-Dability Based On Autonomic Computing

With the constant expansion of the computer system and the growing complexity,the system dependability is a growing concern of network security issues. However,the traditional dependability philosophy or technology do not have the ability of self-management and self-optimization, which seems incapable of action on some possiblenetwork situation and can not meet the dependability requirement in the complex andchanging system. There is an urgent need to find a new approach to improve thesystem dependability without increasing the complexity of the system to make thesystem have the ability of self-optimization in a complex and changing networkenvironment. Autonomic computing which is able to overcome the heterogeneity andcomplexity of computing system has been regarded as a novel and effective approachto implementing self-optimizing to address system dependability issues.The thesis focuses on the challenges of security threats leading to decrease innetwork system service credible performance, trying to resolve the problem of currentsystem dependability models based on autonomic computing and achieve computersystem’s self-management, the dynamic adjustment of resources to accommodate theability of different operating environments. This paper carries out the research aboutthe system dependability model based on autonomic computing and network systemservice performance optimization method which is proposed based on Q-learningalgorithm. The main research work focuses on the following aspects:(1) The paper gives a comprehensive systematic exposition about the basic theoryof the system dependability study and related technologies, and leads the necessary forestablishing the system dependability model based on autonomic computing throughthe analysis of existing technology and methods used in system dependability modelsin the present research.(2) This paper puts forward a system dependability model based on autonomic computing aiming at solving credible performance degradation problem. The modelconsists of three modules: self-control module, the key features of the systemparameters extraction module and self-optimization module. The autonomic controlmodule gives the method to establish self-regulatory unit and the working mechanismof self-discipline management etc. The key features of system parameter extractionmodule gives a key features parameter extraction algorithm using maximum likelihoodestimation based on Bayesian network classification method in order to identifyfactors influencing of the system dependability. The simulation results show that themethod can effectively improve the detection accuracy rate; the self-optimizationmodule designs an autonomous optimization algorithm based on the key characteristicparameters in order to make the system reliable performance be improved.(3) The paper proposed a self-optimization method based on Q-Learningalgorithm. First, service performances are taken as key parameters, which can affectdependability of network system as targets. Second, the three-layer feed-forwardnetwork system is used to map to gain executive action. Finally, the environmentreward function values are calculated according to the change of system serviceperformance and service availability. Then self-learning feature and predictive abilityof Q-learning are used to make the system service performance achieve optimization.