NUMERICAL ANALYSIS METHODS OF SOFTWARE TEST EFFICIENCY

Subject of Research.A nonstationary software testing model is studied. Numerical analysis methods of software testing efficiency based on this model are developed. Modeling of software testing efficiency enables to plan comprehensively the final quality, resources and time required at various project implementation stages. Methods. The technique is based on the proposed improved numerical model for software testing. The process of errors detecting is approximated by the exponential law and the process of elimination by generalized two-phase Cox distribution. The software debugging process after approximation is described by Markovian queue with a discrete set of states and continuous time. The possibility is provided to use the probability of errors detection for each module during their testing. The paper presents the modified marked graph and the system of differential equations; its numerical solution gives the possibility to calculate specific indicators for target effect of software debugging process: probability of certain system states, the time distribution function for errors detection and elimination, the mathematical expectation of random variables and the number of detected or corrected errors. The probability of operating goal achievement (testing) is used as an overall index for the integrated effectiveness evaluating of these processes (including required resources). Main Results. The developed methodology is applied for effectiveness research of the actual project. The private indicators of target effect and integrated efficiency indicator for testing are calculated. The required testing time for specified software quality indicators achievement is identified. The analysis of target effect and time influence on testing effectiveness is performed (on the probability of operation goal achieving). Practical Relevance. The suggested methodology enables to take into account the reliability assessment for each module separately. The Cox approximation removes restrictions on the usage of arbitrary time distribution for fault resolution duration. That generalizes well-known models, simplifies the initial data preparation, improves the accuracy of software test process modeling and helps to take into account the viability (power) of the tests. With these models we can search for the ways of software reliability improvement by generating tests that detect errors with high probability. This methodology gives the possibility to calculate not only the private reliability software indicators, but the integrated indicator of software testing process effectiveness and to develop practical recommendations for effective organization of these processes.

Keywords