Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Studying the Effect of Schedule Preemption on Dynamic Task Graph Scheduling

Mohammadali Khodabandehlou, Jared Coleman, Niranjan Suri, Bhaskar Krishnamachari

TL;DR

Dynamic online scheduling of arriving task graphs on heterogeneous networks is NP-hard; the paper introduces Last-K Preemption to balance adaptability and stability in online DAG scheduling. It formalizes the problem, defines three scheduling paradigms (preemptive, non-preemptive, partially preemptive), and evaluates them across synthetic, RIoTBench, WFCommons, and adversarial workloads. The results show that moderate preemption achieves most of the makespan and utilization gains of full preemption while preserving fairness and lowering runtime overhead, making it practical for IoT/edge settings.

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.

Studying the Effect of Schedule Preemption on Dynamic Task Graph Scheduling

TL;DR

Dynamic online scheduling of arriving task graphs on heterogeneous networks is NP-hard; the paper introduces Last-K Preemption to balance adaptability and stability in online DAG scheduling. It formalizes the problem, defines three scheduling paradigms (preemptive, non-preemptive, partially preemptive), and evaluates them across synthetic, RIoTBench, WFCommons, and adversarial workloads. The results show that moderate preemption achieves most of the makespan and utilization gains of full preemption while preserving fairness and lowering runtime overhead, making it practical for IoT/edge settings.

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.
Paper Structure (25 sections, 4 equations, 8 figures)

This paper contains 25 sections, 4 equations, 8 figures.

Table of Contents

  1. Introduction
  2. Problem Definition & Motivation
  3. Related Work
  4. Scheduling Paradigms
  5. Preemptive Scheduler
  6. Non-Preemptive Scheduler
  7. Partially Preemptive Scheduler
  8. Metrics
  9. Total Makespan
  10. Mean Makespan
  11. Mean Flowtime
  12. Node Utilization
  13. Runtime
  14. Evaluation
  15. Synthetic Task Graphs
  16. ...and 10 more sections

Figures (8)

  • Figure 1: Large leading tasks being blocked by small tasks from previous task graphs
  • Figure 2: Preemptive Scheduler. Having different colors for Scheduled tasks means that they belong to different task graphs.
  • Figure 3: Normalized Makespan
  • Figure 4: Normalized Mean Makespan
  • Figure 5: Normalized Mean Flowtime
  • ...and 3 more figures