Software Aging Trend Prediction Based On Genetic Annealing Neural Network

In a long-running system,software degradation and unplanned outages are called software aging,and software aging is manifested by system state anomalies,performance degradation,software hangs and failures.Several types of systems have been shown to suffer from software aging,including: web servers,operating systems,database management systems,cloud computing,and virtualized environments.Therefore,early prediction of software aging trends and effective precautions before system crashes can help improve software reliability and availability,reduce software failures and unnecessary resource waste and property damage.The paper analyzes the background and significance of software aging research,builds a software aging experiment platform and selects the performance parameters that can reflect the aging trend-average memory,then through comprehensive comparative analysis,BP neural network is selected as the prediction model of this paper.By constructing an experimental platform,a memory leak code was injected into a TPC-W-compliant online book-selling website to age it,and a data set reflecting the aging trend was collected.The average memory of the JVM in 120 seconds was selected and a total of 14281 was collected.Since the original defects of BP neural network will affect the training results,this paper proposes to improve the network structure of BP neural network by improving the genetic algorithm,mainly to improve the selection,crossover and mutation operators of genetic algorithms,and form a new type of self.Based on the genetic algorithm,the simulated annealing algorithm is added to local optimization.The BP neural network model optimized by the proposed adaptive genetic annealing algorithm is systematically described and introduced,and simulated by Matlab.Experiments show that the proposed optimization algorithm utilizes the nonlinearity and adaptability of BP neural network,the parallelism and global convergence of genetic algorithm,the probability of sudden jump and the local convergence of simulated annealing algorithm.Comparison with several traditional improved genetic algorithm optimization BP neural network models for prediction error.The proposed method has significant improvement in prediction accuracy and convergence accuracy,and has great practical value in software aging trend prediction.