TOWARDS TIME-CRITICAL BIG DATA APPLICATIONS IN THE INTERNET OF VEHICLES (IOV)

Abstract

The Internet of Vehicles (IoV) connects vehicles, humans, and infrastructure to produce continual large volume data streams. Real-time processing of this data enables a new class of vehicular applications, however, sending all such events to a centralised cloud will have latency effects that will be unacceptable by many of these applications. To overcome this limitation, this paper suggests a computational offloading model that can process IoV big data in near real time by performing the computation directly on Road Side Unit (RSU) compute resources without the cloud-only processing round-trip delay. The architecture is formalized by utilizing a response-time analysis (RTA) model to determine per node and per cluster utilisation bounds and guarantee that time-critical tasks will meet their response time deadlines under concurrent application load. To test the proposed framework, a real time, location-based advertisement recommendation system was implemented on Apache Kafka and Apache Storm. The inter-RSU travel budget of 3,000 milliseconds was computed for a vehicle travelling at 120 km/h between two RSUs separated by 100 metres. Across 1,000 simulated vehicle events, the combined worst case parsing time and recommendation logic was 1.523 milliseconds, which is well within the 3,000 millisecond inter-RSU travel budget. The results prove that the stream processing can be implemented at RSU without relying on cloud and meet the timing requirement of practical IoV applications.