Latency Estimation and Computational Task Offloading in Vehicular Mobile Edge Computing Applications

Abstract

Mobile edge computing (MEC) is a key enabler of time-critical vehicle-to-everything (V2X) applications. Under MEC, a vehicle has the option to offload computationally intensive tasks to a nearby edge server or to a remote cloud server. Determining where to execute a task necessitates accurate estimation of the end-to-end (E2E) offloading delay. In this paper, we first conduct extensive measurements of the round-trip time (RTT) between a vehicular user and edge/cloud servers. Using these measurements, we present a latency-estimation framework for optimal task offloading. The propagation delay, measured by the RTT, is divided into two components: one that follows a trackable trend (baseline) and the other (residual) that is quasi-random. For the baseline component, we first cluster measured RTTs into several groups, depending on signal strength indicators. For each group, we develop a Long Short-Term Memory (LSTM) regression model. A statistical approach is provided for predicting the residual component, which combines the Epanechnikov Kernel and moving average functions. Predicted propagation delays are incorporated into virtual simulations to estimate the transmission, queuing, and processing delays, hence accounting for the E2E delay. Based on the estimated E2E delay, we design a task offloading scheme that minimizes the offloading latency while maintaining a low packet loss rate. Simulation results show that the proposed offloading strategy can reduce the E2E delay by approximately 60% compared to a random offloading scheme while keeping the packet loss rate below 3%.