engleski

The Topological Data Analysis of Time Series Failure Data in Software Evolution

As evolving complex systems have become the central part of almost every human activity, their reliability is the key property for their successful application, especially with the emerging Internet of Services concept. There are many quantitative mathematical models, the so called reliability growth models, aiming to predict and estimate reliability of software systems based on the failure count time series. This paper suggests a novel and still unexplored qualitative approach to understand failure time series studying its topological features and their influence on failure distributions, thus affecting mission critical system properties, among which is reliability. To illustrate the new ideas, we analyse here the time series failure data of evolving software systems across the system versions for two open source software systems and one mission critical industrial software system, and discuss their topological relations and behaviour. We conclude that topological analysis might be useful for characterising software system behaviour early enough and for early characterization of system reliability that may contribute to software reliability modeling.

Topological data analysis, reliability growth models, software evolution

nije evidentirano