Abstract

This paper deals with a particular class of parallel programs, which are based on task trees. The main objective of this paper was to adapt the generic method for statistical testing of sequential programs (GMST-SP) for this class of parallel programs, such that adapted method (GMST) can treat a family of task trees rather than just a single task tree, and that it can respect various evolutions of individual task trees. In this paper, we compare GMST with the adapted exhaustive testing method (ET) and with the previously adapted statistical usage testing method (SUT), based on experimentally measured testing effort and path coverage. GMST and SUT both have better deep path coverage than ET. SUT requires less testing effort than GMST and ET, but its deep path coverage decreases with the number of tasks. Finally, GMST has advantage over SUT because it provides constant mean level of deep path coverage, which can be regulated by the required testing quality.

Key words: Multicores, parallel programming, parallel programs, structural testing, statistical testing.