Experiences with high-level parallel programming systems

A programmer developing software for parallel systems usually is responsible for the management of low level issues such as communication and synchronization. This is potentially problematic from the perspectives of correctness and portability. This paper considers an alternative approach, namely, the use of high level parallel programming systems. We report our experiences with implementing parallel simulation problem using two very different programming systems: Poker, a high-level tool for parallel programming, based on the programming language C; and Janus, a concurrent logic programming language. Overall, our experience was that the use of such high-level tools greatly simplified the task of developing parallel programs for "ordinary" programmers. While the fundamentally different programming paradigms underlying the two systems resulted in implementations with very different code, our overall conclusions of the strengths and weaknesses of the two systems were remarkably similar. This suggests that the abstractions supported by a high level parallel programming system are fundamentally more important in achieving good programming solutions than the particular underlying programming paradigm.<
>