Computations on the Edge in the Internet of Things

In the Internet of Things (IoT), many applications focus on gathering data which can then be processed and visualized. However, such computations are usually spread generically based on parameters such as CPU and/or network load. This may mean that a significant amount of data needs to be transported over the network (either directly, or transparently using a network file system) in order for the data to be available to the node that is responsible for processing them. This paper proposes a method for deploying computations that can take factors such as data proximity into consideration. Thus, processing can be moved from central high- powered processing nodes to smaller devices on the edge of the network. By doing this, costs for gathering, processing and actuation can be minimized. In order to capture data dependencies among computations, but also to deploy and handle individual processing tasks in an easy way, the actor-model programming paradigm is used. To minimize the overall cost and to handle extra factors that weigh in on the distribution of tasks, a constraint programming approach is used. The combination of these two techniques results in an efficient distribution of tasks to processing resources in IoT. Taking into consideration the NP-hard nature of this problem, we present empirical results that illustrate how this technique performs in relation to the amount of devices/actors.