Parallelizing and optimizing a simulator kernel on a multi-DSP architecture

This paper describes our experiences with the parallelization and optimization of a hydraulic simulator kernel on a multi-DSP platform. Three different parallel versions of the initially sequential simulator have been implemented and optimized with respect to various hardware- and software-related issues. We discuss the paradigms applied for parallelization and the effects certain optimizations have on the simulator's performance. The focus of this research lies in providing measured comparable data of speedup and simulation times for all parallel versions of the same simulator in order to judge the relevance of different parallelization approaches and optimization efforts for a real-time implementation. Selected sets of these data are included and interpreted in this paper.