HUNHODRL: Energy efficient resource distribution in a cloud environment using hybrid optimized deep reinforcement...

Area of Science:

• Cloud Computing
• Artificial Intelligence
• Resource Management

Background:

• Cloud computing resource optimization is crucial for minimizing energy waste and Service Level Agreement (SLA) violations.
• Existing scheduling techniques struggle with dynamic resource allocation, leading to inefficient job completion and container utilization.

Purpose of the Study:

• To propose HUNHODRL, a Deep Reinforcement Learning (DRL)-based framework for advanced container orchestration and workload allocation.
• To enhance resource management efficiency in dynamic cloud environments.

Main Methods:

• Developed HUNHODRL, a DRL framework optimizing scheduling via destination host capacity vectors and active job utilization matrices.
• Evaluated HUNHODRL against HUNDRL, Bi-GGCN, and CNN using CPU, Memory, and Disk I/O utilization metrics across diverse workloads.

Main Results:

• HUNHODRL demonstrated superior performance in container creation rate, job completion rate, and SLA violation reduction compared to existing models.
• Achieved significant improvements in energy efficiency without increasing Virtual Machine (VM) deployment costs.
• Showcased dynamic adaptability to varying workloads, confirming scalability and robustness.

Conclusions:

• HUNHODRL offers a robust and scalable solution for next-generation cloud infrastructure, outperforming traditional methods.
• The DRL-based approach significantly enhances cloud resource utilization and energy-efficient task execution.
• This framework presents a promising direction for intelligent resource allocation in cloud computing.