Font Size: a A A

Research On Key Technologies Of Software Rejuvenation

Posted on:2007-07-19Degree:DoctorType:Dissertation
Country:ChinaCandidate:J XuFull Text:PDF
GTID:1118360215998558Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Computer system becomes more and more complex, and the phenomenon ofsoftware aging resulting from unknown software bugs has come to light. In order toreduce the maintenance cost to recover from the performance degradation and failureof computer system, it is necessary to do research on the key technologies of softwarerejuvenation. In this thesis, the related research background, actuality and problemsare investigated in detail, and several key technologies of software rejuvenation arestudied, including performance metrics selection and monitoring, performancedegradation detection and forecasting. Based on these key technologies, a simplerejuvenation framework is presented. The main contributions of this dissertation aresummarized as follows:Firstly, from the perspective of system resource usage, a impact on systemperformance imposed by the usage of system resource at runtime is explored and acollection of performance metrics is recognized. By improving upon the datacollection, data communication and data storage and so on, a novel performancemonitoring tool is developed, which satisfies the specified monitoring requirementssuch as low latency and low perturbation.Secondly, in order to improve the existing anomaly detection's precision, a novelmethodology inspired by immune identification is presented for performance anomalydetection, which replaces the r-continuous match rule with a new partial match rule innegative selection operator so as to better distinguish between normal and anomaly.The improved negative selection operator is applied to genetic algorithm to act as afilter effectively reduces false alarm rates. The proposed, hybrid algorithm obtainsbetter detection's precision and reduces the whole rejuvenation cost.Thirdly, in order to further reduce rejuvenation cost and improve systemavailability, a hybrid approach is put forward to forecast the optimal rejuvenation time,which combines wavelet analysis and neural network and extends the originalsingle-layer forecast model to be the four stages models. Considering that systemreal-time workload has a great impact on rejuvenation decision-making, the proposedapproach applies semi Markov process to model workload and predicts the systemresource usage and rejuvenation time based on workload model. The experimentalresults show that the proposed algorithm improves the forecasting precision and avoids delaying rejuvenation opportunity.Finally, by analyzing the immune process and its characteristics, simulation andimplement of biological immune mechanism are very valuable to provide a generalsolution framework for software rejuvenation. A simulated rejuvenation architectureof immune mechanism based on multi-agent is put forward. These immune agents areimplemented to monitor performance, detect aging, make rejuvenation decision andtake rejuvenation actions. At last, a simple prototype system for software rejuvenationis proposed, which builds the necessary foundation for further research on softwarerejuvenation.
Keywords/Search Tags:Software Rejuvenation, Software Aging, Performance Monitoring, Performance Anomaly Detection, Artifical Immune System, Time Series Prediction, Software Engineering, Software Architecture, System Availability
PDF Full Text Request
Related items