Computation offloading in MEC-assisted vehicular networks with task migration and result feedback

Abstract

In vehicular networks, roadside units (RSUs) with mobile edge computing (MEC) assistance bring computing resources to the edge for enhancing the computation capacities of vehicles. However, uneven distribution of vehicles leads to load imbalance between MEC computing servers and this poses a huge challenge to computation offloading in vehicular networks. In this paper, we consider task migration between RSUs with different loads on a horizontal scale and computation result feedback in a MEC-assisted computation offloading scenario. We develop a vehicle trajectory prediction module based on deep neural networks for predicting the vehicle position after a task is completed and calculating the delay and energy consumption in the result feedback process. We formulate a computation offloading problem with the aim of minimizing total computation cost within continuous time slots. To address the coupling of decisions under different time slots, we propose a Lyapunov-based novel online heuristic offloading (LNOHO) algorithm for the formulated problem. Applying the Lyapunov optimization framework, the original multi-slot problem is decomposed into multiple per-slot subproblems. Each subproblem is a nonlinear integer programming (NIP) problem. For such an NP-hard problem, we divide it into three processes and propose a load-aware migration heuristic (LMH) algorithm with low complexity to obtain per-slot decisions. The simulation results based on real road topology show that our proposed vehicle trajectory prediction module and algorithm can achieve better performance.