To the issue of predicting the number of errors in the operation of information systems software

Abstract

The assessment of software reliability in information systems is one of the most pressing issues in the modern IT industry. Considering the complexity and importance of tasks performed by information systems, especially systems of critical purpose, the necessity of ensuring their reliability is becoming more and more urgent. In this regard, studying and predicting the residual number of errors that may occur during the operation of the software is extremely important. This article is dedicated to forecasting the residual number of design errors in the software of information systems based on the results of the controlled operation. The paper describes a method for predicting the number of design errors, based on the hypothesis of a random Markov diffusion process with a DN-distribution for time between failures. Although this distribution has been traditionally used as a theoretical reliability model for components, devices, and computer systems, it is a very flexible function of a random argument. The authors of the article suggest that this distribution is worth testing as a model that describes the trend of eliminating accumulated design errors in software that lead to failures. As a result of analyzing thematic publications, a control example of the behavior of some software over time was formed. For this example, its actual failures over a long period of operation are known. The control data were subsequently compared with the obtained theoretical results. The model is described in Python, and calculations were carried out in the corresponding environment. As a result of the simulation, forecast data on the number of failures were obtained using the approach based on the DN-distribution for the time between failures. The evaluation of the results was assessed by the criterion of minimal sum of squared deviations.