Value-Iteration Reinforcement Learning for Distributed Cloud Workflow Scheduling: A Critical-Task-Prioritized Execution Optimization Framework

Authors
  • Chuan He

    New York University, New York, USA

    Author

  • Kuan-Yu Chen

    Northeastern University, Boston, USA

    Author

Abstract

Addressing the complex workflow scheduling problem in distributed cloud environments, this paper proposes a value-iterative reinforcement learning method prioritizing critical tasks. This method addresses the limitations of traditional approaches in handling dynamic resource states, task dependency constraints, and critical task assurance, including insufficient global coverage, limited long-term benefit characterization, and constrained execution optimization capabilities. The method models cloud workflows as task execution structures with dependencies, integrating workflow states, resource states, and dependency information within a unified scheduling framework to construct a state representation mechanism for complex execution processes. Furthermore, it strengthens the priority assurance capability for core task links through criticality evaluation. Building upon this foundation, the method combines value iteration with reinforcement learning decision-making mechanisms to jointly model the immediate benefits and subsequent state evolution during task allocation. This enables the scheduling strategy to achieve global optimization that balances execution efficiency, energy consumption control, resource utilization, and system stability in a multi-node heterogeneous resource environment. Moreover, this paper designs a reward modeling approach that matches the scheduling objectives, allowing critical task delay suppression, resource coordination and allocation, and long-term execution quality improvement to be synergistically driven under a unified goal. The results show that the proposed method has good overall performance in complex cloud scheduling scenarios, effectively improving the execution guarantee capability of critical tasks and the overall scheduling quality, and providing a technical path with application value for intelligent cloud resource management and efficient workflow execution.

Cover Image
Downloads
Published
2025-02-01
Section
Articles