Software and Hardware Redundancy Approaches to Improve Performance and Service Availability in Fog Computing

Abstract

Fog computing is a new distributed computing paradigm. It was introduced to address the massive increase in the number of connected devices since cloud computing faces difficulties handling all requests placed simultaneously. This new paradigm , which is an extension of cloud computing, can increase the efficiency of services provided in many sectors including health care, industry, agriculture, environmental hazard management, smart cities, and autonomous transportation. Some sectors, such as health care and autonomous driving, are highly non-tolerant of delays. In such sectors, high response time and poorly available services can lead to fatal results endangering the lives of many. On the contrary, other sectors such as e-commerce and telecommunication companies can tolerate delays to a certain extent, yet there is always a cost. Delays in such systems do not result in fatalities, as can happen in non-delay tolerant sectors, although delays can cause degraded quality of service and financial loss. Hence, regardless of the level of delay tolerance, delays are not desired. Given the distributed and diverse nature of fog computing, there are some challenges such as device heterogeneity that need be addressed to prepare fog computing for commercial use. Because any device can be a fog node, energy constraints must be considered to maximise device utilisation while still delivering the required quality of service. Also, different devices have various connecting methods which increase complex network connectivity for fog computing. It is also important to consider preventing fog node from exploitation and ensuring that requests are not randomly processed by different fog nodes. This thesis incorporates a management layer in fog computing to address the identified challenges. The proposed model was evaluated using simulations in iFogSim. The results show improved performance in important metrics such as execution time and bandwidth consumption compared to several fog architectures. For higher availability, a duplex management system is proposed and designed using Petri nets. A Markov chain is used to calculate failure probabilities for each node in the management layer, and availability analysis is presented.