Entwurfsraumexploration heterogener Multi-Prozessor-Systeme

Embedded systems are more and more implemented as multi-processor systems. The growing number of applications running in parallel as well as the increasing complexity of the applications result in an increasing number of processors within those systems. Thus, the design space of such systems grows non-linearly. Design space exploration therefore tries to answer the following questions: “How many processors are necessary to fulfill the requirements defined by the application scenario?” and “How can the given application scenario be mapped to the parallel architecture appropriately?” In existing approaches the applications are either manually or automatically mapped to the system. Performance figures are derived either analytically or by running simulations. The results are then used for modifying single applications or improving either architecture or mapping. Due to the complexity of the mapping problem, current approaches following this methodology are too time consuming to permit the exploration of huge design spaces. In this work an approach is presented, that enables global exploration via parallelism analysis. A mapping is not required until the candidate systems identified in the global exploration are inspected in more detail. It is shown that the mapping problem can be regarded as packing problem and solved with linear programming. The comparison of this reference point with the simple heuristic of list-scheduling will show that the latter produces comparatively good results at low computational complexity. This low complexity also allows for using the heuristic approach for dynamic mapping at run-time.