A Reservation-Based Multi-Source Distributed Offloading Strategy in Dynamic Environments

Intelligent Cyber-Physical Transportation Systems (ICTS) provide various services through interaction and integration with vehicles, which can improve the safety and efficiency of the transportation system. This is achieved by leveraging Vehicle Edge Computing (VEC) for processing compute-intensive and latency-sensitive tasks in ICTS. However, most studies make offloading decisions by greedily using all available computing resources in the current network, without fully taking into account the impact of the dynamics of resource occupation on offloading decisions. In this paper, a new Reservation-Based Multi-Source Distributed Offloading (ReMuDO) strategy is proposed. With the goal of minimizing the long-term average task completion latency of the system, this strategy employs a Greedy Randomized Adaptive Search Procedures (GRASP) framework in dealing with the problem of multi-source tasks competing for limited computing resources, and uses the reservation-based Sequential Quadratic Programming (SQP) algorithm to achieve unequal task segmentation to maximize system performance. Extensive experimental results show that the proposed ReMuDO strategy can significantly outperform other classical strategies for different parameters.