Testing Path Generation Algorithm with Network Performance Constraints for Nondeterministic Parallel Programs

Abstract

Nondeterministic behaviors of parallel programs require dedication software testing functionality. It is very difficult to test nondeterministic parallel programs systematically. Although plain reachability testing method can deal with the non-determinism, it may generate huge number of execution paths during the testing procedure. This condition is intractable to cope with in practice. Adding some appropriate constraints to plain reachability testing strategy can improve its efficiency. By analyzing the relationship between message passing behavior and the network performance, the present study proposes a limited reachability testing algorithm for message-passing parallel programs by considering network performance constraints. The number of execution paths is reduced greatly by only considering those execution paths with higher happening possibility according to the network performance constraints. In this way, large-scale parallel programs can be tested more efficiently, and the testing completeness is reasonable.