Dynamic scheduling of task graphs is often addressed without revisiting prior task allocations, with a primary focus on minimizing makespan. We study controlled schedule preemption, introducing the Last-K Preemption model, which selectively reschedules recent task graphs while preserving earlier allocations. Using synthetic, RIoTBench, WFCommons, and adversarial workloads, we compare preemptive, non-preemptive, and partial-preemptive strategies across makespan, fairness, utilization, and runtime. Results show moderate preemption can match most makespan and utilization gains of full preemption while maintaining fairness and low overhead.
@article{arxiv.2602.03081,
title = {Studying the Effect of Schedule Preemption on Dynamic Task Graph Scheduling},
author = {Mohammadali Khodabandehlou and Jared Coleman and Niranjan Suri and Bhaskar Krishnamachari},
journal= {arXiv preprint arXiv:2602.03081},
year = {2026}
}