Energy-Efficient Dynamic Workflow Scheduling in Cloud Environments Using Deep Learning.

Area of Science:

• Cloud Computing
• Artificial Intelligence
• Operations Research

Background:

• Dynamic workflow scheduling in cloud environments is complex due to dependencies, variable workloads, and resource fluctuations.
• Balancing makespan (total completion time) and energy consumption is a key challenge in cloud resource management.

Purpose of the Study:

• To present a novel scheduling framework integrating Graph Neural Networks (GNNs) and Deep Reinforcement Learning (DRL) for multi-objective optimization.
• To minimize makespan and reduce energy consumption in cloud workflows.

Main Methods:

• Utilized GNNs to model task dependencies for adaptive resource allocation.
• Employed Deep Reinforcement Learning with the Proximal Policy Optimization (PPO) algorithm.
• Evaluated the framework in a CloudSim-based simulation environment using synthetic datasets.

Main Results:

• The proposed framework achieved a minimum makespan of 689.22 s, outperforming baseline methods by up to 13.92%.
• Demonstrated consistent improvements in makespan and energy consumption compared to traditional heuristics like HEFT, Min-Min, and Max-Min.
• Maintained competitive energy consumption at 10,964.45 J.

Conclusions:

• The integration of GNNs and DRL offers a powerful approach for dynamic task scheduling in cloud environments.
• The framework effectively balances multiple objectives, including makespan reduction and energy efficiency.
• Findings highlight the potential for advanced AI techniques to optimize cloud resource management.