Keywords: Vehicular Cloud Computing, Task Offloading, Game-Theoretic Optimization, Sustainable Computing
Abstract:
This thesis investigates the technical, economical and environmental feasibility of computing architectures for supporting delaysensitive applications such as Augmented Reality (AR) and Autonomous Driving (AD). While Cloud Computing (CC) is the prevalent computing paradigm today, it cannot offer sufficiently low latency. This limitation is overcome by Edge Computing (EC), which consists in deploying computational capability at the edge of the access network. However, EC entails high infrastructure costs and raises environmental concerns, due to the short lifecycle of electronic devices (around four years) and increased energy consumption.
Meanwhile, the number of connected vehicles is steadily increasing. These vehicles already carry onboard computing and communication resources that can be opportunistically exploited, not only for driving-related tasks, but also for task offloading computation from external devices, e.g., smartphones, laptops, or wearable health devices of end-users. These resources available in the vehicles can be managed under the paradigm of Vehicular Cloud Computing (VCC).
In this thesis, we first analyze the economic feasibility of Edge Computing deployment through a game-theoretic model, showing how multitenant cooperation can mitigate the high cost of deployment. Then, we evaluate under which conditions VCC can replace EC, i.e., whether offloading tasks to vehicles can provide similar performance to EC. Results are obtained via high fidelity network and mobility simulations, in an urban mobile network scenario. We find that VCC can achieve ultra-low latency, with delays of about 10 ms, even when vehicles are sparsely distributed. A comparative cost analysis shows that replacing EC with VCC can reduce infrastructure expenditure by approximately 10% over five years. Finally, we propose a VCC management scheme to optimize energy consumption, carbon emissions, and to compute a fair allocation of the revenues generated by service end-user tasks.
The scheme is based on mathematical programming and coalitional game theory. Via Monte-Carlo simulation, we show that energy consumption due to VCC is below 0.1% of the overall vehicle consumption in realistic scenarios, and that vehicle owners receive substantial incentives for participating in task execution.
In summary, this thesis demonstrates the feasibility of future generation mobile network architectures, such as Edge Computing and Vehicular Cloud Computing, to support extremely low-latency applications.