Abstract
Vehicular cloud computing (VCC) is proposed to effectively utilize and share the computing and storage resources on vehicles. However, due to the mobility of vehicles, the network topology, the wireless channel states and the available computing resources vary rapidly and are difficult to predict. In this work, we develop a learning-based task offloading framework using the multi-armed bandit (MAB) theory, which enables vehicles to learn the potential task offloading performance of its neighboring vehicles with excessive computing resources, namely service vehicles (SeVs), and minimizes the average offloading delay. We propose an adaptive volatile upper confidence bound (AVUCB) algorithm and augment it with load-awareness and occurrence-awareness, by redesigning the utility function of the classic MAB algorithms. The proposed AVUCB algorithm can effectively adapt to the dynamic vehicular environment, balance the tradeoff between exploration and exploitation in the learning process, and converge fast to the optimal SeV with theoretical performance guarantee. Simulations under both synthetic scenario and a realistic highway scenario are carried out, showing that the proposed algorithm achieves close-to-optimal delay performance.
Introduction: The Challenge of Vehicular Cloud Computing
Modern vehicles are increasingly becoming mobile data centers. However, intensive tasks such as AR/VR processing or complex autonomous driving calculations often exceed the local capacity of a single vehicle. Vehicular Cloud Computing (VCC) addresses this by allowing "Task Vehicles" (TaVs) to offload computation to nearby "Service Vehicles" (SeVs) that have spare CPU cycles.
The primary challenge in VCC is the extreme dynamism of the environment. Unlike static edge servers, vehicles move at high speeds, causing wireless channels to fluctuate and the set of available neighbors to change constantly. Traditional centralized offloading schemes require significant signaling overhead to track every vehicle's status. This research proposes a distributed learning approach where vehicles independently learn which neighbors are reliable performers.
The AVUCB Algorithm
The authors frame the offloading decision as a Multi-Armed Bandit (MAB) problem. In this model, each neighboring SeV is treated as an "arm," and the "payout" is a reduction in offloading delay. The core contribution is the Adaptive Volatile Upper Confidence Bound (AVUCB) algorithm, which introduces two key enhancements to the standard UCB approach:
- Occurrence-Awareness: Standard MAB assumes a fixed set of arms. AVUCB handles a "volatile" action space where SeVs appear and disappear. It tracks the "occurrence time" of each SeV to prioritize learning about newcomers.
- Load-Awareness: The algorithm adjusts its exploration strategy based on task size. For small tasks, the cost of a suboptimal decision is low, allowing for "cheap exploration" of less-known SeVs. For large tasks, it prioritizes exploitation of known high-performers to ensure efficiency.
Performance Evaluation
The algorithm was tested in both synthetic and realistic environments. The realistic scenario utilized SUMO (Simulation of Urban MObility) to model a 12km stretch of the G6 Highway in Beijing, including ramps and speeds up to 90 km/h.
As shown in the results, AVUCB consistently outperforms standard learning algorithms like UCB1 and VUCB1 (which is only occurrence-aware). It maintains a delay profile very close to the "Optimal Policy," which assumes perfect future knowledge. Key findings include:
- Regret Reduction: AVUCB reduced "learning regret" by approximately 50% compared to UCB1 in volatile scenarios.
- Fast Convergence: The average delay achieved by AVUCB converged to near-optimal levels significantly faster than baseline methods.
- Theoretical Soundness: The authors provided mathematical proofs that the "regret" grows only logarithmically over time, ensuring long-term performance stability.
Conclusion and Future Directions
The AVUCB algorithm provides a robust, distributed solution for task offloading in highly dynamic vehicular networks. By leveraging task-size-dependent exploration and tracking vehicle availability, it achieves near-optimal performance without heavy communication overhead.
Future work may address the current assumption of independent and identically distributed (i.i.d.) delays, which might not hold under bursty traffic. Additionally, extending the model to support multi-hop offloading and integration with multi-tier architectures (MEC and remote cloud) remains an open area for research.
Summary updated