A Research Of Software Aging Forecasting Method Based On Markov And Neural Network

Software aging means that the performance of a system will be dropped after the software runs a long time. With the continuous development of computer technology, software becomes large part of computer system. The reliability of software has also become a major constraining factor in system reliability. Research and practice shows that long-running software system will be aged inevitably. In order to reduce the harm of software aging, the methods of forecasting and software rejuvenation have been proposed. There are two basic strategies for software aging forecasting-strategy based on time-based and strategy based on measurement.In this thesis, a forecasting method based on Measurement and time has been proposed which also means this method based on Markov and neural network. On the basis of the research of System performance parameters, we can use the method proposed here to judge the changing tend of software state. By this way, a strong basis is provided via the forecasting for more accurate prediction of soft-ware aging. The main content of this thesis are summarized as the following:Firstly, base theories of Markov and artificial neural network are introduced. A kind of BP neural network based on amplified error signal is advanced. It uses BP neural network to forecast the future parameters of the system performance.Secondly, the descriptions of System performance threshold, system state, system state classi-fication, software aging state have been given.Thirdly, in order to collect, analyze and choose parameters of the system performance, a ap-proach is put forward.Fourthly, a forecasting model based on Markov and neural network has been set up. This model trains data by BP neural network and partitions the data by Golden principle. And calculate the range of future values of system parameters, reducing the value of interval, to predict the value of monitor-mg parameters.Finally, experiment is carried out on Linux server, Tool on the server system using Nmon per- formance data were collected. A hybrid approach is used to forecast the key parameter value on the future system, which combines Markov and BP neural network method. Through analyzing of the re-sult of the experiment, we get more accurate value. This provides a strong basis for software aging forecasting.