Failure-detection capability analysis of implementing parallelism in adaptive random testing algorithms

Abstract

Adaptive random testing (ART) has been developed as an enhancement of random testing (RT) in terms of failure-detection capability, and has been widely investigated. When a given faulty program has an N-dimensional input domain (N > 1), a straightforward approach (abbreviated as ART-C) is to implement parallelism into ART algorithms, which generates one N-dimensional test case by computing each of its N coordinates independently and in parallel. Intuitively, ART-C using a specific ART algorithm may not be a great test case selection, because an even spread on each coordinate does not necessarily imply the even spread over the whole input domain. However, actual failure-detection capabilities of implementations of ART-C, so far, have not yet been investigated. In this paper, we conduct on one particular ART algorithm named fixed-size-candidate-set ART (FSCS-ART), and design some simulations to analyze the failure-detection effectiveness of the corresponding implementation of ART-C (that is, FSCS-ART-C). The experimental results show that FSCS-ART-C performs more effectively than FSCS-ART in some scenarios such as high dimensional input domains, which provides useful information for testers to decide whether and when to use FSCS-ART-C. In addition, our study also presents some implications about how to improve the effectiveness of ART.