Efficient deep reinforcement learning based task scheduler in multi cloud environment.

Area of Science:

• Cloud Computing
• Artificial Intelligence
• Operations Research

Background:

• The task scheduling problem (TSP) in cloud computing presents challenges due to dynamic, variable task loads and heterogeneous resources, impacting performance metrics like makespan, energy consumption, and costs.
• Traditional scheduling algorithms struggle with the complexity and NP-hard nature of dynamic cloud workloads.
• Existing metaheuristic and hybrid approaches offer near-optimal solutions but do not fully address the dynamic TSP in cloud environments.

Purpose of the Study:

• To develop an effective task scheduling mechanism for cloud computing environments to address the dynamic task scheduling problem.
• To improve the performance of cloud paradigms by reducing makespan, energy consumption, and resource costs.
• To propose a novel adaptive task scheduler leveraging advanced AI techniques.

Main Methods:

• Formulated an Adaptive Task Scheduler (ATSIA3C) that segments incoming tasks into sub-tasks.
• Modeled the scheduler using an Improved Asynchronous Advantage Actor Critic (IA3C) algorithm for effective task segmentation and scheduling.
• Implemented a two-stage scheduling process: task segmentation and sub-task grouping, followed by VM allocation based on constraints.

Main Results:

• The proposed ATSIA3C demonstrated significant performance improvements over baseline algorithms (RATS-HM, AINN-BPSO, MOABCQ).
• Achieved a 70.49% improvement in makespan, indicating faster task completion.
• Secured a 77.42% improvement in resource cost and a 74.24% reduction in energy consumption in a multi-cloud environment.

Conclusions:

• The ATSIA3C effectively tackles the dynamic task scheduling problem in cloud computing.
• The proposed IA3C-based adaptive scheduler offers a superior solution for optimizing cloud resource utilization and performance.
• ATSIA3C provides substantial enhancements in makespan, cost-efficiency, and energy savings, outperforming existing methods.