PWDCDQ: Prioritized Workflow Scheduling with Deep Reinforcement Learning in Cloud Computing Based on the Deep-Q Network Model

The Cloud computing environments provide scalable resources for complex workflows. Because of the dynamic nature of these environments, resource heterogeneity and varying dependencies of tasks in each workflow differ from one another. Consequently, workflow scheduling with several dependencies on an inappropriate virtual machine, such as one with low processing capacity, can negatively impact the execution, resulting in delays, increased costs, higher energy consumption. To essentially schedule workflows that have numerous task dependencies, we propose a new Prioritized Workflow Scheduling with Deep Reinforcement Learning in Cloud Computing Based on the Deep-Q Network Model (PWDCDQ). To optimize workflow execution, we use a Deep Reinforcement Learning approach, balancing performance with cost optimization by analysing workflows based on their internal dependencies and VMs based on their electricity costs. However, based on both the priorities of tasks, VMs, we employ a DQN model for dynamically scheduling tasks. Essentially, we prioritize tasks based on their dependencies and VMs based on their electricity costs. The priorities are used by the DQN scheduler to assign workflows to the most appropriate VMs, ensuring efficient workflow execution while maintaining low costs. Extensive simulations were conducted on WorkflowSim, examining Real-time scientific workflows, including LIGO, Epigenomics, Montage and CyberShake, are essential tools for conducting research in real-time scientific workflows. The proposed PWDCDQ algorithm was compared to existing approaches such as HEFT, BAT, ACO, and Cuckoo Search (CS) algorithm. The results revealed that PWDCDQ surpasses these approaches in terms of reduced costs, energy consumption.