Heuristically Driven Task Agglomeration in Limited Resource Partially-Reconfigurable Systems

摘要

This paper introduces a method for enhancing run time performance of a dynamic partially reconfigurable system. The technique is applied to fully deterministic task systems that are large in comparison to the resources of the target reconfigurable device. Performance improvements are realized by increasing the granularity of the task system at compile time in a manner that reduces the number of context switches that are required during run-time, thereby decreasing the system execution time. Two algorithms are proposed to implement this technique. Both methods are implemented using simulation, and their performance is compared to a sophisticated heuristic scheduler, which reveals a significant improvement in performance.