مقاله انگلیسی رایگان در مورد محاسبات ابری و مه در اینترنت اشیا – IEEE 2018

In contrast to classical wireless sensor networks (WSN) that usually only serve a single application, one of the core benefits of the shift to the IoT lies in the common usage of sensor hardware by heterogeneous applications (Tschofenig et al., 2015). Additionally, the revolution of the IoT does not stem from the number of connected things alone, but from the solutions and services offered on top of the data. The basic requirements of such value-added services can be briefly summarized into nonvolatile storage of historical sensor data, sensor data processing, and efficient near-real-time distribution of sensor data. However, although everyday objects are increasingly connected to the Internet and becoming more and more powerful, they are usually not capable enough to fulfill all those requirements themselves. One of the major challenges is that commonly used devices, such as sensor nodes, smartphones, and wearable techs, usually run on battery power, making storage or complex processing of a large amount of data unfeasible. Mains-operated connected objects may also often be too constrained to perform those tasks as reliably and quickly as required. The recent advances in cloud computing have led to an increasing usage of this model to meet the aforementioned requirements and enable value-added services in the IoT context. Cloud computing offers convenient, ubiquitous, and on-demand access to a shared pool of configurable computing resources, which are accessible over the Internet, and usually reside in third-party datacenters (Hassan et al., 2012; Mell and Grance, 2011). Along with these resources, cloud providers offer fast and configurable networking for data distribution and reliable, nonvolatile, replicated storage. Thanks to its flexibility, reliability, and usage-based cost model, cloud computing is well positioned to meet the specific requirements of value-added services in the IoT context. Constraint devices can save energy by transferring their data to a cloud-based platform where it will be distributed to multiple relevant applications and services, which will process the data accordingly. Though this architecture works well today, it is not suitable for latencysensitive applications since cloud datacenters are colocated neither with connected objects nor with consumers of value-added services (Zhang et al., 2015). From a network topology view, cloud datacenters are located several hops away from IoT data producers and consumers, and are most often separated by a constrained last-mile link. The physical distance alone causes additional latencies that may not be acceptable for latency-sensitive applications such as control loops. Thus, instead of forcing all IoT communications through a cloud intermediary, there has been a push to move storage, processing, and distribution of data closer to the edge of the network, toward data producers and consumers.