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

Area of Science:

• Computer Science
• Artificial Intelligence
• Cloud Computing

Background:

• Cloud computing task scheduling is NP-complete due to large, dynamic loads.
• Existing methods struggle with efficiency and adaptability in dynamic cloud environments.

Purpose of the Study:

• To propose a novel hierarchical deep reinforcement learning (DRL) framework for large-scale cloud task scheduling.
• To enhance adaptability, cost-efficiency, and performance in dynamic cloud environments.

Main Methods:

• A hierarchical scheduling approach allocating tasks first to VM clusters, then to individual VMs.
• A DRL-based scheduler that continuously learns and adapts network parameters.

Main Results:

• The DRL framework effectively balances cost and performance, optimizing load balancing, cost, and overdue time.
• Achieved a 10% overall improvement compared to classical heuristic algorithms.
• Demonstrated cost reduction in low-load scenarios and improved resource utilization in high-load scenarios.

Conclusions:

• The proposed hierarchical DRL framework offers a promising solution for complex cloud task scheduling challenges.
• Acknowledged limitations include computational overhead, potential latency, and data dependency.
• Further research is needed to address complexity and enhance real-time efficiency.