Managing the far-edge: are today's centralized solutions a good fit?

Abstract

Edge computing has established itself as the foundation for next-generation mobile networks, IT infrastructure, and industrial systems thanks to promised low network latency, computation offloading, and data locality. These properties empower key use-cases like Industry 4.0, Vehicular Communication and Internet of Things. Nowadays implementation of Edge computing is based on extensions to available Cloud computing software tools. While this approach accelerates adoption, it hinders the deployment of the aforementioned use-cases that requires an infrastructure largely more decentralized than Cloud data centers, notably in the far-Edge of the network. In this context, this work aims at: (i) to analyze the differences between Cloud and Edge infrastructures, (ii) to analyze the architecture adopted by the most prominent open-source Edge computing solutions, and (iii) to experimentally evaluate those solutions in terms of scalability and service instantiation time in a medium-size far Edge system. Results show that mainstream Edge solutions require powerful centralized controllers and always-on connectivity, making them unsuitable for highly decentralized scenarios in the far-Edge where stable and high-bandwidth links are not ubiquitous.