摘要
The software reliability testing has many disadvantages in practice, such as high complexity of constructing operational profiles and poor fault detection efficiency. Oppositely, the directed testing with a high fault detection rate is incapable of estimating reliability quantificationally. To solve this problem, a hybrid testing combining reliability and directed testing as well as a reliability model based on the order statistic （OS） model were presented by Mitchell. An extended research on Mitchell＇s work is proposed. Firstly, the most proper distribution of the fault＇s failure rate which tends to be Iognormal is suggested, and a detailed form of the OS model based on Iognormal and the corresponding parameter estimation method are proposed, respectively. Secondly, an im- plementing framework for the hybrid testing is proposed. Finally, the hybrid testing and the OS model are applied on a real website system. The experimental results indicate： the hybrid testing has more efficient fault detection power and lower testing cost than the reliability testing; compared with three traditional software reliabil ity growth models, the OS model has a best or pretty estimation or prediction power for each data set; and for the failure data set collected from hybrid testing, the OS model also achieves an ac- ceptable estimation result.
The software reliability testing has many disadvantages in practice, such as high complexity of constructing operational profiles and poor fault detection efficiency. Oppositely, the directed testing with a high fault detection rate is incapable of estimating reliability quantificationally. To solve this problem, a hybrid testing combining reliability and directed testing as well as a reliability model based on the order statistic （OS） model were presented by Mitchell. An extended research on Mitchell＇s work is proposed. Firstly, the most proper distribution of the fault＇s failure rate which tends to be Iognormal is suggested, and a detailed form of the OS model based on Iognormal and the corresponding parameter estimation method are proposed, respectively. Secondly, an im- plementing framework for the hybrid testing is proposed. Finally, the hybrid testing and the OS model are applied on a real website system. The experimental results indicate： the hybrid testing has more efficient fault detection power and lower testing cost than the reliability testing; compared with three traditional software reliabil ity growth models, the OS model has a best or pretty estimation or prediction power for each data set; and for the failure data set collected from hybrid testing, the OS model also achieves an ac- ceptable estimation result.
作者简介
Corresponding author.Haifeng Li received the bachelor degree in School of Science, the master degree and the Ph.D. degree in School of Reliability and Systems Engineering respectively from Beihang University, China, in 2003, 2006 and 2011. Since 2011, he has been a software reliability testing engineer in the Research Center of Software Engineering of China Aero-Polyteeh- nology Establishment. His research interests include software reliability testing and modeling.E-mail： lihaifeng @ dse.buaa.edu.cnYongql Zhang received the bachelor degree from Nankai University in 1997 and the master degree from Beihang University in 2008. Since 2008, he is a testing engineer in China Network Communication Group. His research interest is software reliability testing. E-mail： zyq@dse.buaa.edu.cnMinyan Lu is a professor and Ph.D. supervisor of School of Reliability and Systems Engineering of Beihang University. She was a visiting scholar at the City University of London from 1994 to 1996. She has published several papers about software reliabil- ity engineering on journals and conferences. Her research interests are software reliability engineering and software dependability. E-Mail： lmy@buaa.edu.cn
