Zum effektiven Einsatz des Adaptiven Zufallstests

Abstract

Software testing, i.e. the systematic execution of the software with the aim of detecting failures, is an essential part of software quality assurance. There are two main problems in software testing. One of them concerns the choice of adequate test data. The other one is the problem of test evaluation. In order to save time and money it is important to automate both the process of test data generation and test evaluation. Random Testing, i.e. the purely random generation of test data within a predetermined input domain, is one strategy of automatically generating test input data. Random Testing data generation is very fast and rather easy to implement. Moreover, its results are unbiased and allow for statistical prediction. Adaptive Random Testing (ART) has been introduced in order to enhance the effectiveness of Random Testing without using any additional information about the software under test. Since there is empirical evidence that failure-causing inputs appear clustered within the input domain, the aim of ART is to achieve an even spread of test cases inside this domain. Many ART methods have been proposed, so far. This work tries to answer the question how ART can be applied effectively. Therefore, previous methods are analyzed and, since they turn out to be inadequate in many situations, new ART methods are proposed.