Faster Fault Finding at Google Using Multi Objective Regression Test Optimisation
Shin Yoo, Rasmus Lind Nilsson, Mark Harman · 2011
Companies such as Google tend to develop products from one continually evolving core of code. Software is neither shipped, nor released in the traditional sense. It is simply made available, with dramatically compressed release cycles regression testing. This large scale rapid release environment creates challenges for the application of regression test optimisation techniques. This paper reports initial results from a partnership between Google and the CREST centre at UCL aimed at transferring techniques from the regression test optimisation literature into industrial practice. The results illustrate the industrial potential for these techniques: regression test time can be reduced by between 33%–82%, while retaining fault detection capability. Our experience also highlights the importance of a multi objective approach: optimising for coverage and time alone is insufficient; we have, at least, to additionally prioritise historical fault revelation. 1.