Semi-Federated Scheduling of Multiple Periodic Real-Time DAGs of Non-Preemptable Tasks

Abstract

Research on real-time scheduling of tasks having different periods has a long history. Passing from uniprocessors to multiprocessors, and from independent sequential tasks to models like DAGs, each was an important turning point causing the need for a heavy mass of new research work. In this research, we consider the problem of scheduling multiple periodic realtime DAGs of non-preemptable tasks on multiprocessor platforms. The DAGs are independent and the deadline of each is until the arrival of the next request for the same DAG. In the context of multiple periodic real-time DAGs scheduling, federated scheduling has been a very successful approach and in this paper we try to extend this idea. The key feature of the federated approach is the dedication of cores to DAGs. An important disadvantage of federated scheduling is that for each DAG a significant portion of its dedicated processing capacity may be wasted. The novelty of the current paper is the proposing of a new method for semi-federated scheduling of multiple DAGs and we try to remedy the mentioned disadvantage to some extent. This is done by compacting the scheduling of each DAG and also relaxing the requirement for scheduling all tasks of a DAG on its dedicated cores. In the experiments section, we compared the proposed semi-federated method with a federated method and significant improvements in success ratio and the number of used cores is achieved, e.g. for a system with 64 cores and a workload of 80 percent, success ratio of the federated method was 0 percent and success ratio of the proposed method was 90 percent.