A Study Of Modeling And Optimizing Software Rejuvenation Policy Based On Non-Periodically Multi-State Inspection

The software aging effect will greatly affect the efficiency of software operation,and even cause system failure,which may lead to catastrophic consequences.To reduce the influence of the aging effect,as a preventive maintenance technology,software rejuvenation has emerged.But due to the overhead of the software rejuvenation behavior itself,the rejuvenation plan cannot be executed frequently.Under the premise of avoiding system failure as much as possible,timely triggering software rejuvenation to minimize system maintenance cost is an important optimization problem.In the traditional inspection-based rejuvenation strategy research,the inspection period is often fixed,that is,the system state is inspected at equal intervals,and then to evaluate whether software rejuvenation is required.Usually,the period inspection mechanism cannot perform "inspection on-demand" according to the system decline law,and there may be unnecessary inspection behaviors in the stable operation stage of the system,but the inspection requires a certain system cost.In addition,the constraints related to the inspection period in the optimization model remain unchanged during the entire system operation,and the constraints are all single-state inspection constraints,that is,the constraints cannot be dynamically adjusted according to the operating state to further reduce maintenance costs.To against the above problems,an aperiodic inspection strategy of multi-state inspection constraint-based and its optimization method are proposed,and the optimization algorithm used is improved.The specific contributions are as follows:1.A multi-state aperiodic inspection strategy in the field of software rejuvenation is proposed.The inspection strategy divides the entire system state into several sub-state intervals,then can independently optimize the Inspection Target Reliability(ITR)--a decision variable related to the inspection period--in different sub-state intervals.The ITR is equal to the probability value that the system will not enter the failure state before the next inspection.It can change dynamically according to the system state,which can further optimize the rejuvenation timing and reduce the maintenance cost of the software system.2.To evaluate and optimize the proposed multi-state aperiodic inspection strategy,a maintenance cost model based on the Markov Regeneration Process(MRP)analyzing method is established.Aiming at minimizing the total system maintenance cost per unit time,based on the Markov Semi-Regeneration Process(MSRP)characteristics of the aperiodic inspection strategy,a discrete maintenance cost evaluation and optimization under the constraint of multi-state ITR is established.Finally,the optimal inspection period will be obtained.By comparing with the traditional fixed inspection cycle and the related maintenance methods of the existing single-state aperiodic inspection cycle,the effectiveness of the proposed method for reducing the system maintenance cost will be proved.3.The exact analytical solution in the optimization model is difficult to obtain,and only the search algorithm can be used.In order to speed up the search algorithm to solve the optimization model,an optimization model solving algorithm based on the fusion of genetic operator improvement is proposed.Firstly,a double-selection operator combining tournament selection and elitism preserving strategy is adopted to avoid excellent individuals not being selected due to randomness or excellent genes from being destroyed due to operations such as crossover and mutation.Secondly,when the genetic algorithm converges slowly in the iterative process,the gradient descent algorithm is used for local search to increase the converge rate.Comparing with the convergence results of the genetic algorithm before the improvement,it will show the effectiveness of the improved method to increase the converge rate.