Energy Efficient Deadline Aware Scientific Workflow Scheduling in IOT- Cloud Environment
U Jeevith, Arunkumar Gopu, Mehul N Sutrave, Kristofel Santa, Norman Baretto · 2025
With growing complexity in cloud computing environments, effective task scheduling has become an essential requirement to optimize resource allocation while maintaining adherence to deadlines. This paper proposes the Adaptive Multi-Agent Reinforcement Learning for Cloud Workflow Optimization (AMARL-CWO) algorithm, where multi-agent reinforcement learning (MARL) is used to optimize cloud resource assignment for scientific workflows. The model is proposed to include three main agents: Task Scheduling Agent (TSA), Resource Management Agent (RMA), and Task Dependency Agent (TDA) that work synergistically to optimize workflow execution. The reward function is designed to balance cost savings, deadline, resource usage, and penalty. The algorithm follows a hierarchical scheduling approach, taking advantage of Directed Acyclic Graphs (DAGs) for task dependency, while utilizing Q-learning and policy gradient for adaptive learning. Also, an adaptive exploration strategy refines scheduling choices with respect to the complexity of workload. Simulation evidence in Cloud Sim shows better performance compared to classical scheduling techniques. The AMARL- CWO algorithm offers an energy- efficient, scalable, and deadline-aware approach to workflow scheduling in IoT-cloud systems, overcoming major challenges of dynamic resource handling. Future improvement involves multi-cloud integration and energy- aware optimizations.