Dynamic adaptation of DAGs with uncertain execution times in heterogeneous computing systems

In this paper, we consider the problem of schedule DAGs with uncertainties in task execution times. Firstly, given an offline planned schedule based on the estimated task execution times, we consider when the schedule should be adapted during runtime based on the current information about the start and completion times of its tasks. The objective is to limit the number of runtime adaptations upon task overruns and underruns and minimize the response time of the DAG. We then consider the case without offline planned schedules and discuss dynamic planning and adaptation of tasks. We conduct extensive simulation experiments to quantify the performance of the proposed algorithms.