Adaptive cloud offloading for vehicular applications

Abstract

The growing number of sensor-based interactive applications and services are pushing the limits of the on-board computing resources in vehicles. With vehicles increasingly being connected to the Internet, offloading the computation to cloud-computing infrastructures is an attractive solution. However, the large sensory data inputs of interactive applications makes offloading challenging across dynamic network conditions, and different application requirements or policies. To address this challenge, we design a system to adaptively offload specific vehicular application components or modules to the cloud. We particularly develop heuristic mechanisms for the placement and scheduling of modules on the On-Board Unit (OBU) and a cloud server under dynamic networking conditions during driving. Through an experimental evaluation of the end-end application response time using our prototype vehicular cloud offloading system, we show that our mechanism can help meet application response time constraints.