Accelerating Regression Testing for Scaled Self-Driving Cars with Lightweight Virtualization -- A Case Study

摘要

Engineering software for smart cyber-physical systems (sCPS) challenges developers as they have to deal with uncertain and volatile stimuli data perceived by sensors. Regression testing of a sCPS is time-consuming on sequential execution. However, sequential testing can be parallelized depending on the system calls used in the system-under-test. In a case study about accelerating regression testing for scaled self-driving cars, we evaluate the use of namespace-separation based lightweight virtualization that powers solutions like Docker or Google's lmctfy. After transparently adding lightweight virtualization to Cxx Test that is used for regression testing, the total test execution time could be reduced from previously over 12min by more than 62% to less than 5min. Thus, the technology for today's lightweight virtualization can also be used to safely accelerate test-runners without changing existing test cases.