Toward Service Offloading in Cloud-Fog Ecosystem Using Hybrid Approach of Transfer and Deep Reinforcement Learning

Fog computing extends the ability of the cloud and integrating these two provides a viable alternative in order to improve the Quality of Service (QoS). The integrated cloud-fog environment is dynamic and varies depending upon network conditions, resource availability, and workload requirements. A framework is warranted that automatically adapts to such dynamic changes and efficiently deploys the limited computing fog nodes for millions of applications by preserving QoS. A two-phase service offloading framework is proposed in this research that comprises the classification of the services followed by scheduling of services. An ML-based decision tree is implemented to classify the services. For scheduling, we propose the multi-agent-based modified twin delayed deep deterministic algorithm (TD3), a hybrid of TD3 and Transfer learning (TL), i.e., TD3-TL. The transfer learning is introduced to reduce the computation complexity of model. The proposed model has been simulated on Google trace data, and its performance is measured on various QoS, where it yields better results than a few existing models.