Edge Computing-Enabled Internet of Vehicles: Towards Federated Learning Empowered Scheduling
Feng Sun, Zhenjiang Zhang, Sherali Zeadally, Guangjie Han, Shiyuan Tong
Abstract
Classical edge computing algorithms assume that the execution time is always known in resource allocation. However, in practice, the execution time in the edge server is hard to estimate due to the complex environment, especially in Internet of Vehicles (IoV), which makes resource allocation a significant challenge. To address this problem, we propose an optimal resource allocation approach based on Federated Learning (FL). In our proposed approach, we consider both the delay and energy consumption. First, we assume that we have perfect knowledge about the execution time in the edge server, and we obtain the optimal CPU cycles which should be assigned to process 1-bit task. We also verify the Poisson property of the task arrival for each server pool in the edge server. Since we use Delay Energy Product (DEP) as our optimization target for resource allocation, we can derive the optimal balanced delay and energy performance in the scheduling. Next, we propose the Federated Learning based method to estimate the execution time in the edge server. Simulation results obtained demonstrate the effectiveness of our proposed approach.