Adaptive Container Migration Strategy for Delay-Sensitive and Dependent Tasks in Internet of Vehicles

The integration of edge computing and container technologies offers new opportunities for the advancement of Internet of Vehicles (IoV). However, existing research lacks of the consideration of task complexity and variability of traffic flow in IoV. In this paper, we propose an adaptive container migration strategy tailored for latency-sensitive and interdependent tasks. Specifically, we consider the dependency relationships among in-vehicle tasks and the mapping between tasks and containers. We formulate an optimization problem as a Mixed-Integer Linear Programming (MILP) to optimize task completion rates and system load balancing. Building on this, we propose an adaptive container migration strategy by RIME algorithm and a hierarchical task scheduling algorithm with deadlock avoidance, where migration decisions are made based on dynamic changes in traffic flow, allowing better adaptation to fluctuating traffic demands. We introduce a Bayesian optimization approach to determine the optimal migration threshold. We validate the proposed solution using real-world datasets. Experimental results demonstrate that, compared to other algorithms, our approach significantly enhances task completion rates and server load balancing. Moreover, it can effectively strike a balance between system performance and migration frequency.