Research On Intelligent Fine-grained Software Rejuvenation Policy And Related Techniques

As the complexity of computer system increases, many software bugs inducing software aging cannot be avoided, it has been a problem that we have to face, so software rejuvenation is introduced. In order to advance the efficiency of software rejuvenation and reduce the cost of rejuvenation and the loss that results from performance degradation and failure of computer system, the intelligent and fine-grained software rejuvenation should be researched urgently. For these requirements of application, the main contributions of this dissertation are summarized as follows:Firstly, the new determination method based on software engineering for the degree of restart dependence between components and restart tree of software system is put forward, which aims at the problem of roughness. By analying software system architecture and coupling relation between components, the dataed expression of coupling relation between components is presented; then the degree of restart dependence between components is obtained by searching accessible path, on which the restart group of any module is defined. So the primary preparation for software rejuvenation finishes".Secondly, the multi-level nested software rejuvenation policy is modeled and analyzed. In this paper, both Petri net and finite-state automata are used to depict the k-level nested rejuvenation policy. In the process of executing a policy, Petri net is used to denote dynamic behavior of every restart in detail, and automata is used to control the restart components, the sequences and implement periods of them. The obtained policy model is simple and understandable, and this method avoids the blast of state space that would occur if the model was solved directly by Markov process.Thirdly, the intelligent fine-grained software rejuvenation policy of distributed system is established, and the architecture of agent systems is built. Based on the function characteristics of components on different levels, the four -level nested rejuvenation policy is set down and rejuvenation granularity is fined to thread process level. By analyzing the coupling relation between components on different levels, the methods to search the direct coupled components and the restart sets are put forward; by inspecting the usage of system resource, rejuvenation components and the nested restart relations are illuminated, the restart tree and restart chain are formed, and the archetechture is formed; so the intelligent fine-grained software rejuvenation policy of distributed system is grounded . Fourthl y, through analyzing the realization of intelligent software rejuvenation, the principium and model are set down. The model of artificial neural network,genetic algorithm and DNA computing based on software rejuvenation are built and determined. By analyzing the function principiums of artificial neural network,genetic algorithm and DNA algorithm, the availbilities of them are determined; on the basis of principium of software rejuvenation, the model of five-level artificial neural network is built, and genetic algorithm and DNA computing are set down, so the degree of restart dependence and reachable sets of components are get intelligently. So that the efficiency of software rejuvenation is improved, the cost decreases, and the necessary surport of intelligent software rejuvenation is provided.Fifthly, the model of data streams system based on software rejuvenation is built, so that thecharacters of real time and veracity of policy are enhanced. On the basis of data streams and software rejuvenation principium, by determination of general performance features, extraction and trend detection of unusual patterns of data streams, the model of data streams system based on software rejuvenation is built, which determines the degree of restart dependence between components, the restart ones and sequences of them, then the restart chain of system is built. The efficient and veracious software rejuvenation realized, and the rejuvenation effect is improved.