Computing power network dynamic resource scheduling integrating time series mixing dynamic state estimation and...

Area of Science:

• Cloud Computing
• Artificial Intelligence
• Resource Management

Background:

• Multi-cloud architectures present challenges in cross-cloud resource scheduling, including adapting to heterogeneous environments and slow dynamic load responses.
• Optimizing resource utilization while maintaining service quality is a critical issue in cloud management.

Purpose of the Study:

• To propose an intelligent scheduling framework, TSL-HRL, that addresses the challenges of multi-cloud resource scheduling.
• To improve resource utilization efficiency and scheduling real-time performance in dynamic multi-cloud environments.

Main Methods:

• The TSL-HRL framework integrates time-series feature modeling and hierarchical reinforcement learning.
• A time-series mixing module extracts features from resource fluctuations and demands.
• A dynamic state estimation module with Kalman filtering captures real-time resource supply and demand.
• A hierarchical reinforcement learning architecture with Q-learning (high-level) and A2C (low-level) algorithms enables collaborative decision-making.

Main Results:

• TSL-HRL effectively balanced resource utilization efficiency and scheduling real-time performance in experiments.
• The framework demonstrated robust performance on the Multi-Cloud Service Composition Dataset and Google 2019 Cluster dynamic node scenarios.
• The three-level architecture design proved effective for time-series feature extraction, dynamic state perception, and hierarchical strategy optimization.

Conclusions:

• TSL-HRL offers a systematic solution for resource management in multi-cloud environments.
• Future work includes extensions for edge-cloud collaboration, energy optimization, and meta-learning for rapid adaptation.