Multi-Hop Task Routing in Vehicle-Assisted Collaborative Edge Computing

• Published in: IEEE transactions on vehicular technology Vol. 73; no. 2; pp. 2444 - 2455
• Main Authors: Deng, Yiqin, Zhang, Haixia, Chen, Xianhao, Fang, Yuguang
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Collaboration; Collaborative edge computing; Computation offloading; Deep reinforcement learning; deep reinforcement learning (DRL); Edge computing; Mobile computing; multi-hop routing; Multiagent systems; Optimization; Relays; Resource management; Resource utilization; Routing; Routing (telecommunications); Servers; Spread spectrum communication; Transportation networks; Vehicular ad hoc networks; vehicular networks
• ISSN: 0018-9545; 1939-9359; 1939-9359
• DOI: 10.1109/TVT.2023.3312142

Collaborative edge computing has emerged as a novel paradigm that allows edge servers (ESs) to share data and computing resources, effectively mitigating network congestion in traditional multi-access edge computing (MEC) scenarios. However, existing research in collaborative edge computing often limits offloading to only one hop, which may lead to suboptimal computing resource sharing due to challenges such as poor channel conditions or high computing workload at ESs located just one hop away. To address this limitation and enable more efficient computing resource utilization, we propose a multi-hop MEC approach that leverages omnipresent vehicles in urban areas to create a data transportation network for task delivery. Here, we propose a general multi-hop task offloading framework for vehicle-assisted collaborative edge computing where tasks from users can be offloaded to powerful ESs via potentially multi-hop transmissions. Under the proposed framework, we formulate an aggregated service throughput maximization problem by designing the task routing path subject to end-to-end latency requirements, spectrum, and computing resources. To efficiently address the curse of dimensionality problem due to vehicular mobility and channel variability, we develop a deep reinforcement learning, i.e., multi-agent deep deterministic policy gradient, based multi-hop task routing approach. Numerical results demonstrate that the proposed algorithm outperforms existing benchmark schemes.