DAU Library

학위논문 동서

學習現場과 誤診發見率을 고려한 소프트웨어의 最適放出政策

김인환

부산: 東亞大學校 1995. -

56p: 삽도; 26cm. -

학위논문(석사)- 東亞大學校 大學院 産業工學科 1995년12월

영문초록 : A research field of software reliability is to predict quantitatively software reliability based on Software Reliability Growth Model(SRGM) which has contributed to evaluation for developed software systems. The previous research assumed that debugging process is perfect with probability one. But it is generally assumed that debugging process is imperfect because it involves imperfect human factor and the error detection rate is not only affected by content of remaining error content, but also affected by other factors. Therefore this research proposes two plausible software reliability growth models based on nonhomogeneous Poisson process(NHPP) with learning factor for imperfect debugging, that is, the debugging process improves with experience, and the time-dependent behavior of the error detection rate. The first model is formulated based on NHPP with the quadratic learning function of debugging rate differently from the linearly increasing learning factor of Xia et al. But it can't solve closed form of mean value function in this model, and it is difficult to use it in determining optimal release policy. The second model is developed by NHPP with linearly increasing learning factor and error detection rate represented by Rayleigh function. Actual software failure data are applied and compared with the results of Xia et al. Optimal software release policies, based on the proposed model with infinite life cycle length and random life cycle length, respectively are studied. The price of the software system and three cost components are considered expected total profit and software reliability requirement are used as the criteria. Also, a method of determining optimal release time, in the case that cost factor itself contains conventional learning phenomenon and time dependent pattern of error detection rate are proposed. Numerical examples on these optimal software release policies are presented for illustration. Further problems remain to investigate the form of error detection rate which have diverse forms according to type of softwares, and methods to derive an exact mean value function under the intricated state of circumstances. And though more superior in terms of MSE than Xia's model, this model use more parameters that. So, Future researches about the scope and number of adequate parameters may be necessary.

바로가기