Enhancing Adaptive Random Testing through Partitioning by Edge and Centre Tsong Yueh Chen Fei-Ching Kuo Huai Liu ∗ Faculty of Information and Communication Technologies Swinburne University of Technology Hawthorn Victoria 3122, Australia {tchen, dkuo, hliu}@ict.swin.edu.au Abstract Random Testing (RT) is a simple but widely used software testing method. Recently, an approach namely Adaptive Random Testing (ART) was proposed to en- hance the fault-detection effectiveness of RT. The ba- sic principle of ART is to enforce random test cases as evenly spread over the input domain as possible. A va- riety of ART methods have been proposed, and some research has been conducted to compare them. It was found that some ART methods have a preference of se- lecting test cases from edges of the input domain over from the centre. As a result, these methods may not per- form very well under some situations. In this paper, we propose an approach to alleviating the edge pref- erence. We also conducted some simulations and the results confirm that our new approach can improve the effectiveness of these ART methods. 1. Introduction Random Testing (RT) is a fundamental software test- ing method, which selects test cases randomly from the input domain (the set of all possible inputs of the program under test) [11, 18]. Due to its advantages, such as simplicity, efficiency and randomness, RT has been popularly applied in industry. For example, RT has been used to examine the reliability of some UNIX utility programs [16, 17], and it was found that a great number of programs were crashed by random test data. Moreover, RT was also applied in many automatic test- ing tools, such as those developed by Microsoft [19], IBM [2] and Bell Lab [10]. It has been observed that failure-causing inputs (pro- gram inputs that can reveal failures) tend to cluster to- gether [1, 3, 9]. Some researchers [8, 14] found that under such a situation, the performance of RT can ∗ Corresponding author be significantly enhanced by evenly spreading gener- ated test cases over the whole input domain. This ap- proach was named as Adaptive Random Testing (ART). Based on their work, many ART methods have been proposed, and some typical examples include Fixed- Sized-Candidate-Set ART (FSCS-ART) [8, 14], Re- stricted Random Testing (RRT) [4], and Lattice-based ART [15]. These methods have been experimentally evaluated and it was confirmed that ART can use fewer test cases to detect the first failure than RT when failure- causing inputs are clustered into contiguous regions (namely failure regions [1]). Recently, some research has been conducted to com- pare some ART methods [12], and it was pointed out that although all ART methods have the same aim, that is, evenly spreading test cases, their performances are different from one another because they distribute test cases using different approaches. One of the most im- portant observations of the research is that FSCS-ART and RRT, which provide the best performances when failure rate is small, prefer to select test cases from the edge part of the input domain rather than from the cen- tral part. This preference, however, deteriorates the per- formances of FSCS-ART and RRT as the failure rate and dimension increase, and it even can make FSCS- ART and RRT less effective than the original RT under some situations [6, 7, 12]. In this paper, we study the edge preference of a particular ART method, namely FSCS-ART. We pro- pose a new approach to alleviating the preference, and thus to enhancing the performance of FSCS-ART. The structure of the paper is introduced as follows. Sec- tion 2 gives some background information of FSCS- ART. Section 3 introduces how we measure the edge preference of a testing method, explains our approach to alleviating the edge preference, and compares the fault- detection effectiveness of our approach and the original FSCS-ART. Finally, Section 4concludes the paper. Proceedings of the 2007 Australian Software Engineering Conference (ASWEC'07) 0-7695-2778-7/07 $20.00 © 2007 Authorized licensed use limited to: SWINBURNE UNIV OF TECHNOLOGY. Downloaded on March 24,2010 at 01:44:03 EDT from IEEE Xplore. Restrictions apply.