Studying the Effect of Schedule Preemption on Dynamic Task Graph Scheduling

2025-10-06

IEEE Military Communications Conference

AI Summary

This work addresses the challenge of balancing performance and fairness in dynamic task graph scheduling, where traditional approaches often neglect adjustments to existing task assignments. To overcome this limitation, the authors propose the Last-K Preemption model, which introduces a controlled, localized preemption mechanism that reschedules only the most recent K task graphs while preserving earlier allocations. This strategy effectively balances scheduling efficiency against system overhead. Extensive experiments are conducted using synthetic, RIoTBench, WFCommons, and adversarial workloads, comparing fully preemptive, non-preemptive, and partially preemptive strategies. The results demonstrate that the proposed moderate preemption approach achieves makespan and resource utilization comparable to full preemption, while significantly reducing scheduling overhead and ensuring fairness.

Technology Category

• Planning, Routing, and Scheduling: Scheduling under Uncertainty
• Game Theory and Economic Paradigms: Fair Division
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems

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/utilization gains of full preemption while maintaining fairness and low overhead.

Problem

• schedule preemption
• dynamic task graph scheduling
• makespan
• fairness
• resource utilization

Innovation

• schedule preemption
• dynamic task graph scheduling
• Last-K Preemption
• makespan optimization
• resource utilization