A pilot study to compare programming effort for two parallel programming models

We evaluate the claim that a PRAM-like parallel programming model (XMTC) requires less effort than a message-passing model (MPI), through a quasi-experiment conducted with students in graduate-level parallel programming courses. Our main result is that XMTC programs required about 45% less effort than MPI programs. There was insufficient power to detect a statistically significant difference in the rate of correctness between the two models. These results suggest that if architectures continue to evolve towards fine-grained uniform-memory-access parallel machines, XMTC-like languages are worth pursuing. However, further studies are necessary to evaluate this claim with respect to different types of problems, as well as to larger programs. While the sample size of this study is smaller than we would have liked, obtaining subjects for such studies is difficult. The population of programmers who have training in parallel programming is small, so we rely on available parallel programming courses for novice subjects. Nevertheless, we feel that this type of study is a good first step in the continued empirical research of parallel programming issues, and provides a basis of comparison for future studies which may involve more experienced programmers and different programming tasks.