Multi-objective task scheduling using SBA-based deep reinforcement learning in cloud computing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cloud Computing

Background:

• Cloud computing offers scalability but faces challenges in efficient resource management.
• Timely task execution and high availability are critical for cloud services.
• Optimizing resource allocation is key to reducing operational costs and energy consumption.

Purpose of the Study:

• To propose a novel task scheduling approach for cloud environments.
• To improve resource allocation, utilization, cost-efficiency, and energy consumption.
• To enhance high availability, particularly for long-term jobs.

Main Methods:

• A hybrid approach combining a Social-Based Algorithm (SBA) with Deep Reinforcement Learning (DRL), termed SBA-DRL.
• Task allocation based on learning workload patterns and adapting to characteristics in batch scheduling.
• Evaluation using a synthetic dataset and the real-world Google Cloud Jobs (GoCJ) dataset.

Main Results:

• SBA-DRL reduced costs by up to 28.94% and energy consumption by up to 25.31%.
• Resource utilization improved by up to 14.04% across both datasets.
• Outperformed existing scheduling methods in resource allocation and high-availability management.

Conclusions:

• The SBA-DRL approach effectively addresses cloud task scheduling challenges.
• Offers a practical solution for enhancing the efficiency and sustainability of cloud systems.
• Demonstrates significant improvements in cost reduction, energy efficiency, and resource utilization.