Improved Conditionally Optimal DAG Task Parallelization for Global EDF

The DAG task model has gained significant attention and widespread usage for modeling complex real-time applications, particularly in domains like autonomous driving. To effectively utilize these models, research on the schedulability of DAG tasks has been catalyzed. Notably, a recent study has focused on selecting the optimal parallelization option for scheduling in a multi-core, global EDF environment. However, we have identified limitations in this study that hinder its full potential for improving schedulability, especially in a heterogeneous task model, where the parameters of the tasks (e.g., execution time, period, and deadline) vary. In such task model, the selection of a node set for parallelization can impact the schedulability dramatically. To this extent, in this paper, we propose an advanced approach to the parallelization selection method, that employs the following two strategies: 1) fine-grained consideration of the ‘tolerance benefit’ to select the next optimal node and 2) determine a tighter bound for the interfering workload, that greatly reduces over-estimation of the interference. Through extensive simulation experiments, we demonstrate that the proposed methods yield significant improvements in schedulability.