Reinforcement learning based multi objective task scheduling for energy efficient and cost effective cloud edge...

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• The proliferation of Internet of Things (IoT) devices necessitates efficient task scheduling in hybrid cloud-edge environments.
• Traditional scheduling algorithms struggle with dynamic workloads and conflicting objectives like performance, energy, and cost.
• Latency-sensitive applications require optimized resource allocation for timely execution.

Purpose of the Study:

• To introduce Reinforcement Learning-Based Multi-Objective Task Scheduling (RL-MOTS) for intelligent resource allocation.
• To develop a framework that balances task latency, energy consumption, and operational costs.
• To enhance the adaptability and scalability of task scheduling in heterogeneous cloud-edge systems.

Main Methods:

• Formulating task scheduling as a Markov Decision Process.
• Utilizing Deep Q-Networks (DQNs) for adaptive resource allocation.
• Implementing a priority-aware dynamic queueing mechanism and a multi-objective reward function.
• Employing a state-reward tensor for real-time decision-making across heterogeneous nodes.

Main Results:

• RL-MOTS achieved up to 28% reduction in energy consumption and 20% improvement in cost efficiency.
• Significant reductions in makespan and deadline violations were observed compared to baseline strategies.
• The framework maintained strict Quality of Service (QoS) requirements under varying workload conditions.
• Demonstrated adaptability to both preemptive and non-preemptive scheduling scenarios.

Conclusions:

• RL-MOTS offers a sustainable, cost-efficient, and performance-oriented solution for next-generation distributed computing.
• The framework provides intelligent and adaptive resource allocation for hybrid cloud-edge environments.
• Future work will explore transfer and federated learning for enhanced scalability and privacy in decentralized systems.