Supporting Data Stream Analytical Processing in Vehicular Sensor Networks

Abstract

In the past decade, the Vehicular Sensor Network (VSN) technology has emerged as a promising technique to support Intelligent Transportation Systems (ITSs). Utilizing the information collection and communication capabilities provided by VSN, data can be firstly collected by in-vehicle sensors, and then uploaded to the infrastructure to support ITS applications. However, to get a global view of the status of the road, many ITS applications adopt a centralized approach, which requires to collect data from VSNs to a central server. In addition, to provide timely services, data has to be updated by vehicles continuously. As a result, massive amount of data could be generated by vehicles and transmitted in the network, which may exhaust the limited wireless communication bandwidth. In this work, we propose a data stream analytical processing framework for VSN named Streaming Vehdoop (SVehdoop). To reduce the bandwidth consumption, SVehdoop schedules part of data processing tasks to where the data is located. Specifically, SVehdoop utilizes the computing capability of vehicles to efficiently process the collected data over a large number of vehicles in a distributed manner. A dynamic clustering algorithm, named Streaming Vehdoop Clustering (SVC) algorithm, is tailor-designed for SVehdoop to not only consider vehicle mobility to form stable clusters, but also to take account of data aggregation and data parallelism. Comprehensive experiments have been conducted to demonstrate the efficiency of SVehdoop and the proposed SVC algorithm.