DEEP NEURAL NETWORKS ADAPTATION IN FOG COMPUTING FOR EFFECTIVE DATA ANALYTICS

Abstract

Fog computing focus on local processing of tasks on edge Internet of Things (IoT) devices known as fog devices. Nowadays IoT applications produce a large amount of data and require powerful analytical approaches. These data should transmit to cloud data centres to extract useful information. In fog computing, these kinds of data handling by fog layer. The fog layer consists of geo-allocated servers which are deployed on the network periphery. Each fog server can be known as a lightweight version of the cloud server. In particular, it is preferred to the expand fog server, which reduces data amount before sending them to cloud data centres by using TensorFlow, one of the most popular deep learning library, Google Cloud Platform and Deep Neural Network. Deep neural networks can be known as a computer system modelled on the human brain and neural system. In data classification, neural network provides fast and efficient results. In this scenario fog server known as fog station and focused on implement analytical applications in fog stations with the help of Docker and Kubernetes, except implementing them in the main cloud server. Those applications belong to one or more fog devices. Through the research work, studied and compared the effects of using TensorFlow, with multiple hidden layers. TensorFlow library with the help of keras used to build the neural network model. The experiment results of Rectified Linear Unit (ReLu), Leaky ReLU, Hyperbolic Tangent (tanH), Exponential Linear Unit (eLu), sigmoid, softplus, softmax and softsign activation functions have been evaluated. The experiment results demonstrated that the feasibility, efficiency and the applicability of the proposed fog station. Finally, achieved (i) Centralized Management using Fog Station, (ii) Dynamic Deployment using Docker, (iii) Efficient Management & Resource Monitoring using Kubernetes, (iv) Real-Time Data Analytics using TensorFlow and GCP as objectives.