Studying the Effect of Schedule Preemption on Dynamic Task Graph Scheduling

TL;DR

This paper investigates controlled schedule preemption in dynamic task graph scheduling, proposing the Last-K Preemption model to improve scheduling efficiency while balancing fairness and overhead.

Contribution

It introduces the Last-K Preemption model, a novel approach that selectively reschedules recent tasks to optimize performance and fairness.

Findings

Moderate preemption achieves similar makespan and utilization as full preemption.

Preemptive strategies improve scheduling metrics with low overhead.

Partial preemption balances efficiency and fairness effectively.

Abstract

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.