Elastic and Energy-Efficient Virtual Machine Allocation on Edge Servers for Internet of Vehicles

Edge computing and virtualization are two important technologies to support future Internet of Vehicles (IoV). Edge servers deployed at Road-Side Units (RSUs) employ Virtual Machines (VMs) to provision realtime, close-distance and low-latency task offloading services to the vehicles on road. Nevertheless, the spatiotemporal traffic dynamics introduce a great challenge in managing the VMs. The local traffic surge can overwhelm the VM capacity of a server, and the VMs have to be frequently powered on and off to catch up with the traffic change, which consumes a great amount of energy. To address these issues, we propose an elastic and energy-efficient VM allocation strategy on edge servers for IoV systems. First, we conduct a trace study on a real traffic trajectory dataset, which reveals the pervasive complementary traffic patterns across neighboring regions. Such insight inspires us to propose an elastic VM allocation strategy by expanding the service range of an edge server to its neighbors. We build a mathematical model of the VM allocation problem, and design an online heuristic algorithm to solve it based on the temporal difference between the current number of VMs and the traffic load in the following round in each region. Extensive trace-driven simulations verify that our algorithm can achieve an up to 35.1%-reduction of the energy consumption compared with other popular approaches, while ensuring that all the vehicles are served by the VMs.