基于Logistic测试覆盖率函数的软件可靠性建模研究

Software Reliability Modeling with Logistic Test Coverage Function

软件测试覆盖率是测试充分性和测试效率的有效度量指标,其与软件可靠性以及缺陷覆盖情况之间有着一定的相关关系,并且结合测试覆盖率信息的软件可靠性模型的评估和预计效果将会得到有效改进.在实际测试过程中,由于软件结构特征及学习因素的综合影响,测试覆盖率可能会呈现出一种先增后减的趋势,Logistic函数恰好非常适合描述这类S形变化趋势,且结构简单,具有较好的灵活性与适应性.因此,针对基于Logistic函数的测试覆盖率函数以及软件可靠性建模等问题展开研究.首先提出基于Logistic函数的测试覆盖率函数;在该函数的基础上,提出基于Logistic测试覆盖函数的缺陷预计模型;然后,将NHPP可靠性模型的建模过程与Logistic测试覆盖函数相结合,提出一种新的考虑测试覆盖率的软件可靠性增长模型.实例验证结果表明:与若干已有的同类研究成果相比,提出的基于Logistic函数的测试覆盖率函数、缺陷预计模型以及软件可靠性增长模型有效地提高了函数或模型对数据的拟和精度,且具有较好的适用性.

Test coverage is a good indicator for testing completeness and effectiveness,which has some effects on software reliability and defect coverage.Furthermore,the evaluation results of software reliability models with test coverage information will be further improved.Although some of traditional software reliability growth models have been widely applied to reliability prediction,it is likely that the prediction accuracy of these models can be further improved by adding other important factors affecting the final software reliability.Test coverage is believed to be one of such factors.Due to the integrated effects of software structure and learning factor on testing,test coverage increasing rate may exhibit a varying trend which first increases and then decreases.In other words,test coverage function may be an S-shaped curve which can be well and flexibly described by the Logistic function in many situations.Hence,this paper utilizes the Logistic function to describe the test coverage growth behavior.Based on this Logistic test coverage function,a defect prediction model that relates test coverage to fault detection is presented.Furthermore,combining NHPP modeling framework with Logistic test coverage function,a new software reliability model considering test coverage is proposed.Several case studies are also presented and analyzed.The results indicate that compared with several existing models,the new defect prediction model and reliability model considering Logistic test coverage function proposed in this paper can provide a substantial improvement in term of goodness-of-fit power and applicability.