DaRkaM : um framework de redução de dados baseado na névoa aplicado ao contexto da computação urbana

Abstract

The Intelligent Transport Systems (ITS) has a function to analyze the flow of vehicles on highways in order to identify any traffic anomalies, ensuring greater efficiency during the decision-making process. These systems can be based on an Ad-Hoc vehicular network (VANETs) able to integrate the elements of urban space through a distributed commu- nication system. Similarly, ITS applications require constant monitoring of roads, and such monitoring aims to analyze, among other aspects, the variation of vehicular density over time. In general, this process occurs by means of the periodic sending of situational data from the mobility environment to the cloud. Consequently, the data sets are sent at high frequency to the cloud, forming a continuous data flow that should be proces- sed in a real-time context. However, this dynamic implies a progressive increase in the communication cost, in reason of the volume of data transferred in the link between the fog and the cloud, increasing the risks of overload beyond increasing the latency during requests for services made available in the cloud. Therefore, this work proposes the de- velopment of a multilayer architecture for data reduction based on Fog Computing called DaRkaM, acronym in English for (Data Reduction Framework for Traffic Management ). The strategy consists of using a monitoring model able to perform data reduction pro- cesses directly at the edge of the vehicular network. At the cloud layer, DaRkaM acts as a central controller, analyzing the geographic positions of vehicles that are received from a continuous data flow. These data are used to monitor and perform the traffic manage- ment processes addressed in this proposal. At the edge network, a data reduction module was designed to host different traffic monitoring strategies. This architecture favors the comparative analysis among different approaches, ranging from the use of algorithms ba- sed on simple sampling to clustering algorithms, in which the data reduction processes are structured based on the number of clusters. The results showed that the use of cluster- based algorithms, hosted in the data reduction core of the DaRkaM framework, is able to achieve a great reduction in communication cost, especially in overloaded scenarios, beyond achieving high accuracy in monitoring and detecting traffic congestion.