Abstract

The Internet of Things (IoT) infrastructure forms a gigantic network of interconnected and interacting devices. This infrastructure involves a new generation of service delivery models, more advanced data management and policy schemes, sophisticated data analytics tools, and effective decision making applications. IoT technology brings automation to a new level wherein nodes can communicate and make autonomous decisions in the absence of human interventions. IoT enabled solutions generate and process enormous volumes of heterogeneous data exchanged among billions of nodes. This results in Big Data congestion, data management, storage issues and various inefficiencies. Fog Computing aims at solving the issues with data management as it includes intelligent computational components and storage closer to the data sources. Often, an IoT-enabled infrastructure is shared among many users with various requirements. Sharing resources, sharing operational costs and collective decision making (consensus) among many stakeholders is frequently neglected. This research addresses an essential requirement for adaptive, autonomous and consensus-based Fog computational solutions which are able to support distributed and in-network schemes and policies. These network schemes and policies need to meet the requirements of many users. In this work, innovative consensus-based computational solutions are investigated. These proposed solutions aim to correlate and organise data for effective management and decision making in Fog. Instead of individual decision making, the algorithms aim to aggregate several decisions into a consensus decision representing a collective agreement, benefiting from the individuals variant knowledge and meeting multiple stakeholders requirements. In order to validate the proposed solutions, hybrid research methodology is involved that includes the design of a test-bed and the execution of several experiments. In order to investigate the effectiveness of the paradigm, three experiments were designed and validated. Real-life sensor data and synthetic statistical data was collected, processed and analysed. Bayesian Machine Learning models and Analytics were used to consolidate the design and evaluate the performance of the algorithms. In the Fog environment, the first scenario tests the Aggregation by Distribution algorithm. The solution contribute in achieving a notable efficiency of data delivery obtained with a minimal loss in precision. The second scenario validates the merits of the approach in predicting the activities of high mobility IoT applications. The third scenario tests the applications related to smart home IoT. All proposed Consensus algorithms use statistical analysis to support effective decision making in Fog and enable data aggregation for optimal storage, data transmission, processing and analytics. The final results of all experiments showed that all the implemented consensus approaches surpass the individual ones in different performance terms. Formal results also showed that the paradigm is a good fit in many IoT environments and can be suitable for different scenarios when applying data analysis to correlate data with the design. Finally, the design demonstrates that Fog Computing can compete with Cloud Computing in terms of accuracy with an added preference of locality.