Coordinating Computation at the Edge: a Decentralized, Self-Organizing, Spatial Approach

Fog and Mobile Edge Computing are promising paradigms aimed at bringing cloud-like functionality at the edge of the network, close to end users and IoT devices, hence complementing the offer of traditional cloud computing, which is based on powerful but distant data centers. A notable, specific thread of research explores so-called edge-clouds, i.e., small clouds emerging from the combination of resources of proximate edge-devices, whose goal is to provide on-demand storage and computation power to nearby users. In order to support the engineering of such edge computing ecosystems, in this paper we describe an approach for coordinating resources and computations in edge-clouds that assumes connectivity only to nearby devices (abstracting from the concrete communication technology) and tolerates unreliability by self-adaptation to device failure, mobility and withdrawal. Most notably, we delineate a decentralized, self-organizing, spatial approach that works by dynamically partitioning the system into areas, each one governed by an elected manager, and setting up downstream and upstream coordination flows from managers to peripheral nodes (i.e., workers and users) and vice versa. We provide an implementation schema in the ScaFi framework for aggregate programming and evaluate a basic request scheduling scenario through simulation.