Dependency Aware Workflow Scheduling mechanism in cloud computing by using deep deterministic policy gradient algorithm

Workflow scheduling in cloud – fog computing environments is a challenging multi-objective optimization problem due to task dependencies, heterogeneous resources, and dynamically varying workloads. Existing heuristic and deep reinforcement learning (DRL)-based scheduling approaches often exhibit limited adaptability in integrating dependency-aware task prioritization with real-time resource allocation and reliability-aware scheduling decisions. To address these limitations, this study proposes an adaptive workflow scheduling framework that integrates Random Forest Regression (RFR)-based task priority estimation with a Deep Deterministic Policy Gradient (DDPG)-based continuous-action scheduling mechanism. The proposed framework models scientific and industrial workflows as Directed Acyclic Graphs (DAGs), where workflow dependencies are incorporated into both task prioritization and scheduling decisions. A hybrid priority estimation mechanism combining upward-rank heuristics and data-driven task ranking is developed to improve dependency-aware scheduling efficiency. Furthermore, the DDPG scheduler dynamically considers virtual machine (VM) availability, workload conditions, trust values, execution history, and reliability indicators during task allocation. A multi-objective reward formulation is designed to jointly optimize workflow makespan, resource utilization, energy consumption, trust, and reliability. The proposed framework was implemented using a custom discrete-event simulator developed on top of the SimPy environment and evaluated using Google Cluster workload traces across small-scale (100–350 tasks), medium-scale (400–650 tasks), and large-scale (700–1000 tasks) workflow scenarios. Experimental results demonstrate that the proposed framework reduces workflow makespan by 10.1%, improves resource utilization to 94.3%, reduces energy consumption by 13.4%, and improves scheduling reliability by 8.7% compared with existing DRL-based approaches including DQN, A2C, and A3C. The results confirm that integrating dependency-aware task prioritization with continuous-action reinforcement learning significantly improves adaptive workflow scheduling performance in dynamic cloud-fog computing environments.