Research On Software Aging Trend Prediction Of Web System Based On Recurrent Neural Network

With the maturity of Internet technology,the expansion of user groups and the increase in demand,the number of software is increasing and the functional structure is becoming more and more complex.The more convenient and reliable requirements make the software maintenance more difficult.However,the long-running software system will inevitably lead to performance degradation and system crash due to the aging phenomenon,which seriously threatens and restricts system's high reliability and availability to easily cause incalculable losses.Although software rejuvenation can eliminate the adverse effects of software aging,however,when to implement rejuvenation is a question.Over-frequent regeneration can easily lead to an increase in downtime and further loss.Too low regeneration frequency can not solve the aging problem in time.If the trend of software aging can be predicted accurately,the phenomenon of software aging will be found in time before system failure.A reasonable software regeneration strategy can be adopted to solve the problem of system performance degradation so that the loss caused by software regeneration can be reduced as far as possible,the adverse effects caused by downtime can be effectively avoided,and the reliability of the system can be improved.Therefore,how to predict the aging trend of the software accurately is a key problem of preventing software aging.Aiming at the problem that the current aging trend prediction method does not consider the dependence and the low accuracy of aging data,this paper builds a model of memory resource consumption in Web system based on the deep learning recurrent neural network,combining with the characteristics of mutation and multi-dependence of aging data.Firstly,according to the temporal characteristics of aging data,this paper presents a Web system resource prediction method based on recurrent neural network(RNN)and its variant–Long Short-term Memory(LSTM),whose network structure is optimized.Secondly,aiming at the problem of low utilization of long-distance information and different influences on the final trend of LSTM network,a deep neural network resource consumption prediction model based on self-attention mechanism(SATLSTM)combing with LSTM network and the self-attention mechanism is proposed.Finally,Using system memory usage as a measure of aging,the two aging prediction models are modelled and predicted by building a software aging test platform due to memory leak and using an accelerated aging test experiment.Comparing withexisting prediction models,the above predicting model is vertified the effectiveness.The experimental results show that the aging prediction model based on recurrent neural network can effectively capture the long-term dependence of aging data,and proves to be superior to the other traditional models with the predicted results closer to the actual values and the higher prediction accuracy so that this model is more suitable for describing the process of software aging resource consumption.The improved SATLSTM aging prediction model has further improved the prediction accuracy,and more accurately describes the hidden rules of aging performance parameters,which lays a theoretical foundation for the identification and regeneration of software aging state.