Software Test Resource Allocation Based On Adaptive Operator Selection

With the rapid development of computer software,the complexity of software systems has increased dramatically.In order to ensure the reliability of software systems,the consumption of software testing resources has inevitably increased.Under the condition of limited resources,how to find the balance between resource consumption and reliability has obtained more attention.Optimization of software test resource allocation not only involves the establishment of software reliability and cost models but also to seek an effective balance between software reliability,test resource consumption,and software release time.Therefore,optimizing software test resource allocation is a multi-objective optimization problem.Based on the analysis of the existing software reliability growth model,this paper finds that when modeling a software reliability model people prefer to simply omits the process of fault correction.To solve this problem,we adopt a software reliability growth model with fault correction in this paper.In the past,the optimization of software testing resource allocation was to minimize the testing cost and maximize the reliability under the condition of limited test resources.In this paper,we add the third optimization goal of minimizing the software release cycle and we use the difference evolutionary algorithm based on adaptive operator and neighborhood size selection to solves this optimization problem.The following are the major improvements in this article:(1)It has been found that the widely used non-homogeneous Poisson process software reliability growth model relies entirely on the assumption that a potential fault is detected and then immediately it is removed.In the actual test process,there is certainly a time delay between the fault being found and being corrected,and it is possible that the fault correction lags behind the fault detection seriously.To solve this problem,this paper adopt a Software Reliability Growth Model Based on Time Delay(BTD-SRGM).Experiments show that the BTD-SRGM model can betterly fit the actual software testing process.(2)A Least Squares based on Evolutionary Algorithm(LSEA)is proposed.Generally,scholars will find the first-order partial derivatives of the least-squares when estimating the parameters of reliability growth model.Such an approach does not guarantee a global optimal solution,the most likely outcome is a local optimal solution,this will bring great bias to the experimental results.In this paper,Based on the good global convergence of the evolutionary algorithm,we apply the evolutionary algorithm to the least-squares,And comparative experiment is made.(3)An new adaptive operator selection evolutionary algorithm based on Multi-Armed Bandit(MAB)is proposed.In the iterative process of evolutionary algorithm,there is no any operator or neighborhood size can be well applied to the entire search stage.However,many evolutionary algorithms have the disadvantages of single operator and neighboring size in the search process.To solve this problem,we propose an algorithm based on Multiarmed Bandits theory to decide which operator to be selected,and to improve the utilization of the well-behaved operator through the attenuation mechanism.Considering the diversity of Pareto front,the distance-sorting algorithm is embedded in the proposed algorithm then we get our new algorithm MAB-MODED.Experiments show that MAB-MODED achieves acceptable software reliability with a shorter testing cycle and a lower testing cost.