Дослідження точності оцінювання надійності програмного забезпечення

Abstract

Наведено залежність точності оцінювання надійності програмного забезпечення від мови програмування і моделі надійності. Виконано моделювання надійності для декількох програмних систем за допомогою основних моделей. Встановлено, що точність оцінювання надійності програмного забезпечення не залежить від мови програмування.

Приведена зависимость точности оценки надежности программного обеспечения от языка програмированния и модели надежности. Выполнено моделирование надежности для нескольких программных систем с помощью основных моделей. Показано, что точность оценивания надежности программного обеспечения не зависит от языка программирования.

The aim of the work is to establish the presence or absence of dependence of the accuracy of reliability assessment on the programming language and software reliability model. To this end, software reliability modeling was performed using the main reliability models, such as: Dzhelinsky-Moranda, non-uniform Poisson process (Gela-Okumoto), Schneide-Windows, Musa, Weibul model, S-Shaped model, Duena, geometric model of Moranda, Musa-Okumoto. The existence of the problem of choosing a reliability model, which is due to their large number, is noted. It is shown that the problem of choosing a model has not yet been resolved. For research, we selected time series for defect detection in 40 software systems written in various programming languages: JavaScript, Ruby, Python, Objective-C, C ++, Scala, C #, PHP, C, Java, Rust, ActionScript. The data source for the specified time series is the Internet resource Github.com. Modeling was carried out using specialized software developed by the authors. The simulation accuracy was estimated as the mean-squared deviation of the calculated cumulative defect detection curve from the real one. The dependence of the accuracy of software reliability assessment on the programming language and reliability model is given. Recommendations are given on choosing a model for a software system depending on the programming language. It is concluded that there is no one universal model that with acceptable accuracy would allow us to evaluate the reliability of a software system, regardless of the programming language in which it was written.