Non-work conserving dynamic scheduling of moldable gang tasks on multicore systems

In the realm of cyber-physical systems, the prevalence of multicore/many-core systems equipped with multiple or numerous cores is evident. To efficiently handle these systems, multicore task scheduling techniques are imperative. Particularly in recent cyber-physical systems, the dynamism of these systems renders task scheduling a complex endeavor. In this paper, we introduce a dynamic scheduling algorithm for moldable gang tasks on multicore systems. Our proposed algorithm determines when and with how many cores a task should be executed, especially for tasks with uncertain arrival times. In our experiments, we evaluate whether our algorithm reduces the makespan compared to the latest dynamic scheduling algorithms for moldable gang tasks. We demonstrate that, given a sufficient number of cores relative to the number of incoming tasks, our algorithm can reduce the makespan by approximately 10%.