Distributing Computations in Fog Architectures

Abstract

Internet of Things (IoT) services and applications include Connected Vehicles, Smart Grid, Smart Cities, Health Care and, in general, Wireless Sensors and Actuators Networks. Typically, the scenarios can be captured with a Fog Computing architecture that consists of edge nodes that generate and possibly pre-process (sensor) data, fog nodes that do some processing quickly and do any actuations that may be needed, and cloud nodes that may perform further detailed analysis for long-term and archival purposes. This paradigm enables (i) quicker real time computations and actuations, avoiding the latency involved in communicating with the cloud for them, (ii) reducing the amount of data that is sent to the cloud, thus reducing network bandwidth requirement and delay in data transmission, and (iii) doing this without the need for 24/7 network connectivity to the cloud. However, the storage, compute and network connectivity capabilities of the edge and fog nodes may be limited. Hence the computations need to be distributed carefully among the processing nodes. In this paper, we develop a generic framework for distributing computations to the different nodes in a fog architecture. Our framework is applicable to an arbitrary hierarchy of the nodes, one or more homogeneous or heterogeneous source inputs, and to processing the input batches either individually or combined with other batches by way of merges and splits. It can serve initially as a schema for a given computation and later to optimize executions of instances.