Machine learning based computa tional offloading in the Fog Cloud Internet of Things ecosystems

Abstract

Internet of Things (IoT) is the new prototype that is shaping Information and Communication Technologies (ICT) in a new direction. The rate of adaptability, ingenuity, and expansion is promising to bring connectivity to all things in homes, offices, vehicles, etc. This paradigm is causing researchers to find new ways to connect things efficiently, improve processing power of devices, create value for data collected from IoTs, and ensure the security of the data. Although numer ous data-intensive applications (self-parking cars, trackers, domestic appliances, smartphones, etc.) have been developed to employ the use of smart devices they are small in size, battery-powered, and limited in processing, storage, and memory supplies. Their size and nature of applications act as a bottleneck to implement powerful applications on them. Moreover, the amount of data generated by IoT de vices collectively surge the network. Therefore, it is necessary to utilize the Cloud of Things (CoT) infrastructure. A CoT provides extensive processing power and unlimited storage that permits fast processing, and bulk storage of data produced by the new cloud infrastructure. It works well in situations that are not delay sensitive, and require no immediate responsiveness. However, the use of CoTs exclusively is not effective in applications that require immediate processing, high responsiveness, and real-time analysis of client’s requests because of the distance between the CoTs and IoTs. To this purpose, fog computing has been designed. Fogging enables computational offloading, data aggregation, and storage. The offloading process enables smart objects to realize the full potential of the IoT Fog-cloud infrastructure by sending part of its computation routines to remote sites for processing. This permit saving computational resources such as proces sors, storage, memory, and energy. Nonetheless, the question of when, what, and how to offload has not been fully resolved. Moreover, new challenges keep emerg ing such as finding the better offloading point given mobility, heterogeneity in IoT devices, and dynamic communication environment. Further, dealing with offload ing that is constrained by location and latency-sensitive submissions remain hard to solve. This study is intended to solve an IoT-Fog infrastructure performance problem in which each of the IoT devices may contain computation-intensive, or security-sensitive, or delay-sensitive tasks to offload. If the IoT device finds that it cannot execute the tasks, an offload to an optimal Fog is initiated through a Smart gateway (SG). The Fog either performs the tasks or sends it to the cloud. The in tension of the study is to perform dynamic offloading while maintaining the user’s sensitive tasks in the Fog during offloading. The study is expected to achieve high performance in terms of throughput, delay, energy consumption, resource utiliza tion rate and response time. For this purpose, a computation offloading framework is proposed with the engagement of a four-tier cloud infrastructure using a pipeline of machine learning (ML) strategies. This pipeline consists of a set of ML models connected in series to facilitate the offloading efficiently. The study is intended to design a fast, efficient and robust algorithm that enables the selection of optimal fog and cloud when the need arises, while providing mobility when network con ditions fluctuate.