Parallel Dynamic Test Generation

Software testing is an crucial step in software quality assurance. In the process of software testing, developers always try to achieve high code coverage in order to uncover potential software defects. Practically, providing suitable test inputs usually leads to good coverage, so that a hot topic in the research of software testing is how to automatically generate expected test inputs. To achieve this goal, some researchers proposed dynamic symbolic execution, a program analysis technique which is applied to generate test inputs. Dynamic symbolic execution technique executes the program under test concretely, and collects symbolized path constraints as well as concrete program states within the concrete program executions in order to find out suitable test inputs for specific program paths. However, this dynamic technique faces performance challenges when programs under test become more and more complex, especially when there are loops in target programs which cause infinite execution paths. On the other hand, in recent years people have changed the way to improve computation efficiency, from raising CPU frequency to adding extra computing units and introducing parallel algorithms.This paper describes the parallel dynamic symbolic execution technique for automatically and systematically generating test inputs. By this parallel technique, the exploration of program paths can be distributed to different computing units in order to boost data generating performance. The implementation of this approach and corresponding experiment results show that parallel dynamic symbolic execution is very practical for the real testing environment.