The emergence of federated cloud edge computing has brought challenging issues in scheduling dynamic tasks, which require the consideration of energy efficiency, latency, SLA satisfaction and migration of resources together. Primitive approaches such as ru le based and heuristic scheduling can be too rigid and unable to deal with the volatile and diverse behavior of contemporary distributed systems. Although recent deep reinforcement learning (DRL) methods have achieved favorable performance, most existing m ethods possess issues such as hard reward specification and a lack of support for multi objective optimization and federated scalability/privacy. To fill these gaps, we propose FedTaskRL in this paper, a new federated DRL based DT scheduling framework for a cloud edge ecosystem. The proposed model employs a neural Q learning algorithm with an augmented state representation and a raw multi objective reward function. Such a design allows the model to adaptively learn customized scheduling policies to cut down energy, reduce response time, comply with SLA, and save migration cost over time for federated learning while considering the data locality of these clients. We conduct thorough extensive experiments in a federated simulated setting and show that FedTask RL outperforms state of the art methods, including DRL TS, A3C Scheduler, DRLIS, EdgeTimer, and MA DRL. The designed framework has a 28 kWh Use of Energy, 145 ms Average Response Time, 97. 5% of SLA fulfilment, and a lower cost of 4. 8 for the migration. T hese findings also confirm the effectiveness and efficiency of FedTaskRL in real time cloud workload management. In summary, FedTaskRL provides a scalable, adaptive, and privacy respecting solution for intelligent task scheduling, resulting in significantl y improved performance and practicality in federated cloud edge resource management.
A 2025 study studied this question.