An Algorithm for Static Tracing of Message Passing Interface Programs Using Data Flow Analysis

Abstract

Message Passing Interface (MPI) is a well know paradigm that is widely used in coding explicit parallel programs. MPI programs exchange data among parallel processes using communication routines. Program execution trace depends on the way that its processes are communicated together. For the same program, there are a lot of processes transitions states that may appear due to the nondeterministic features of parallel execution. In this paper we present a new algorithm that statically generates the execution trace of a given MPI program using data flow analysis technique. The performance of the proposed algorithm is evaluated and compared with that of two heuristic techniques that use a random and genetic algorithm approaches to generate trace sequences. The results show that the proposed algorithm scales well with the program size and avoids the problem of processes state explosion which the other techniques suffer from.