Research On Modeling And Evaluation Of Autonomic Dependable System

As software complexity and malicious attacks continue increasing, dependability has now become one of the most serious challenges, but the traditional dependability which depends on designing more complex software to be dependable is not suitable. In this paper, based on the feature of self-awareness and self-management, autonomic computing is used to design a new kind of automatic, flexible and low complexity dependability, autonomic dependability. We are tempted to explore a new angle to dependability research by studying the model and evaluation attributes of Autonomic Dependable system (ADs).Firstly, the relationships between the concept and attributes of autonomic dependability and the traditional dependability are analyzed, and then the correspondence relationship between these two groups of attributes is set up. Secondly, how to build the framework of the modules for ADs such as self-tolerance, self-recovery, and self-destruction modules is discussed, and owing to it, a self-reflection model based on theπ-Calculus is constructed. Thirdly, a state model for ADs based on a Markov process is presented on the foundation of analysis and abstract of the architecture of ADs.Researches on the evaluation attributes of ADs are helpful to determine the main factors to dependability, decrease vulnerabilities, and improve dependability. According to the definition and quantization of the metrics, the evaluation metrics architecture is constructed, and then an evaluation model based on Analytic Hierarchy Process (AHP) is built. In addition, an autonomic dependability evaluation tool, ADE-Tool, is developed to meet the requirements of dynamic and large-scale evaluationAt last, the proposed model of ADs based on a Markov process is solved to get the steady-state probability, and its stability is analyzed by sensitivity analysis. Besides, in order to test the ADE-Tool, the self-healing attribute of an experiments system is evaluated.