A heterogeneous computing system provides a variety of differentmachines, orchestrated to perform an application whose subtasks havediverse execution requirements. The subtasks must be assigned tomachines (matching) and ordered for execution (scheduling) such that theoverall application execution time is minimized. A new dynamic mapping(matching and scheduling) heuristic called the hybrid remapper ispresented here. The hybrid remapper is based on a centralized policy andimproves a statically, obtained initial matching and scheduling byremapping to reduce the overall execution time. The remapping isnon-preemptive and the execution of the hybrid remapper can beoverlapped with the execution of the subtasks. During applicationexecution, the hybrid remapper uses run-time values for the subtaskcompletion times and machine availability times whenever possible.Therefore, the hybrid remapper bases its decisions on a mixture ofrun-time and expected values. The potential of the hybrid remapper toimprove the performance of initial static mappings is demonstrated usingsimulation studies
展开▼