Research On Software Aging State Recognition Method Based On Optimized GBDT-LR Model Under Weibull Distribution

Software aging refers to the phenomenon of gradual decline in performance,abnormal status,instability or even sudden failure in the process of software operation.Software rejuvenation is an active maintenance technology aimed at software aging.It recognizes the aging state of the software,and restarts the software system after determining that the software enters the aging state to clear the internal error state and error parameters of the software system,thereby improving the overall The purpose of software system performance.Since restarting will cause the software system to temporarily stop service and cause loss of time and performance,the accurate identification of software aging status has always been one of the key issues in the field of software aging and rejuvenation research.At present,the main research on software aging and rejuvenation is focused on time-based(model)rejuvenation methods and monitoring-based rejuvenation methods.The former performs mathematical modeling of the software running process and calculates the rejuvenation(restart)operation at a certain time interval,but it lacks the ability to grasp the real-time situation of the software system and the ability to self-adjust.The latter monitors the real-time hardware parameters related to the operation of the software system,predicts the trend of the hardware parameters according to the machine learning algorithm,and uses this as a basis to identify the aging state of the software system.However,it only considers a single resource occupancy and hardware status factors,and lacks relevance to the aging status of different software systems and adaptability to demand conditions.Because the two methods are used alone,there are certain shortcomings.Based on the idea of ??the model-monitoring hybrid method,this paper proposes a software aging state recognition method.This paper innovatively introduces the classic Weibull distribution in the field of reliability analysis to simulate the state transition process of software system aging,and then combines real-time hardware parameters and calculated model parameters to optimize the weighting of AHP coefficients,and uses GBDT-LR hybrid model performs the final software aging state recognition operation.The method proposed in this paper solves the problem of system flexibility and adaptability caused by the use of a single model method,and balances the unpredictable factors in software aging in the process of using the mathematical model,thereby solving the use of a single machine learning method to a certain extent The resulting lack of correlation between hardware parameters and software aging phenomenon.The experimental results show that the method proposed in this paper performs better than the traditional solutions using a single model or a single monitoring algorithm in terms of the accuracy of software aging status recognition,time consumption,average task processing speed,and system stability.