Towards an efficient data analytics architecture for the Internet of Things

Abstract

In the Internet of Things (IoT), the traditional architecture aims to process the data in the cloud. This creates several challenges such as high communication latency between the end devices and the cloud while making the network busy by sending all the raw data continuously. In this thesis, we propose an alternative architecture for the IoT which processes part of the data in the fog to avoid all raw data to be sent to the cloud. However, the cloud processes intensive data analytics. We conduct a trade-off analysis to show the advantages of applying data fusion closer to the data source and then processing the intensive data analytics algorithms in the cloud. We explore the effectiveness of the available architectures including centralised, decentralised, and distributed architecture to propose the most effective data analytics architecture for the IoT. The trade-off analysis shows the effectiveness of various service decomposition strategies leading to an understanding the various balances between Fog and IoT processing and their effectiveness in data communications reduction and result accuracy allowing achievements of 70% data communication reduction while still achieving approximately 90% accuracy. We propose a service distribution strategy called Most Efficient IoT Node (MEIN), which aims to distribute the services to either cloud nodes or fog nodes based on their capabilities while maintaining the usage of resource in IoT architecture. This strategy selects the best nodes and distributes the services on nodes based on the demands of services and capabilities of nodes.