IoT traffic management using deep learning based on osmotic cloud to edge computing

IoT is critical in many application areas, such as smart cities, health care, and surveillance systems. Each application has its own QoS requirements. Dynamic traffic management in an IoT network is essential for optimal load balancing and routing. It also allows applications to achieve their desired level of QoS. Osmotic computing is a paradigm for edge/cloud integration. In this paradigm, to balance the load of the network hosts, the services must migrate from a higher resource-utilized data center to a smaller one. According to the osmotic computing approach, each IoT application could be broken into some Micro-Elements (MELs), and each MEL resides on a resource on the edge or cloud data center. Usually, in an IoT osmotic environment, services must be executed by the edge hosts. Some remaining services must migrate to the cloud data centers if the edge hosts lack computational resources. Therefore, such data migration may produce massive traffic across the network. Moreover, the traffic sometimes must pass through a particular route, which includes some pre-specified nodes, for security or monitoring reasons. The routes must be optimized regarding QoS metrics such as delay, jitter, and packet loss ratio. Therefore, finding an optimal path between the source and the destination MEL is essential. Deep learning can facilitate this process by exploiting the massive routing data to find the optimal routes with pre-specified node(s). For this purpose, this paper proposes a new traffic management algorithm based on a deep RNN model. The algorithm predicts the alternative optimal routes, including the desired node (s), in an IoT osmotic environment. A collection of paths is generated using the minimum-distance maximum-bandwidth routing algorithm to create the dataset. The IoT osmotic environment consists of three main layers: the edge data center, Software-Defined Wide Area Network (SDWAN) infrastructure, and cloud data centers. The proposed traffic management algorithm is implemented in the controller of each layer. The simulation results showed that the osmotic approach increased the energy consumption of the edge devices and reduced the transaction time. Because the data is processed near the user, the flow size of the traffic, which is sent across the network, is reduced. The experimental results also showed that the model could achieve up to 94% accuracy. The model training and prediction time do not affect the application's total running time.