An novel cloud task scheduling framework using hierarchical deep reinforcement learning for cloud computing

Delong Cui, Zhiping Peng, Kaibin Li, Qirui Li, Jieguang He and Xiangwu Deng

PLOS ONE, 2025, vol. 20, issue 8, 1-25

Abstract: With the increasing popularity of cloud computing services, their large and dynamic load characteristics have rendered task scheduling an NP-complete problem.To address the problem of large-scale task scheduling in a cloud computing environment, this paper proposes a novel cloud task scheduling framework using hierarchical deep reinforcement learning (DRL) to address the challenges of large-scale task scheduling in cloud computing. The framework defines a set of virtual machines (VMs) as a VM cluster and employs hierarchical scheduling to allocate tasks first to the cluster and then to individual VMs. The scheduler, designed using DRL, adapts to dynamic changes in the cloud environments by continuously learning and updating network parameters. Experiments demonstrate that it skillfully balances cost and performance. In low-load situations, costs are reduced by using low-cost nodes within the Service Level Agreement (SLA) range; in high-load situations, resource utilization is improved through load balancing. Compared with classical heuristic algorithms, it effectively optimizes load balancing, cost, and overdue time, achieving a 10% overall improvement. The experimental results demonstrate that this approach effectively balances cost and performance, optimizing objectives such as load balance, cost, and overdue time. One potential shortcoming of the proposed hierarchical deep reinforcement learning (DRL) framework for cloud task scheduling is its complexity and computational overhead. Implementing and maintaining a DRL-based scheduler requires significant computational resources and expertise in machine learning. There are still shortcomings in the method used in this study. First, the continuous learning and updating of network parameters might introduce latency, which could impact real-time task scheduling efficiency. Furthermore, the framework’s performance heavily depends on the quality and quantity of training data, which might be challenging to obtain and maintain in a dynamic cloud environment.

Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0329669 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 29669&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0329669

DOI: 10.1371/journal.pone.0329669

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().