A Deep Reinforcement Learning-Based Approach to Resilient Service Placement for Mobile Edge Computing

In the Internet of Vehicles (IoV), Mobile Edge Computing (MEC) alleviates the computational and storage limitations of vehicle terminals by deploying cloud computing resources at the network edge, substantially reducing service access latency. However, the dynamic nature of vehicular networks introduces significant challenges to service placement strategies due to the tension between fluid network topologies and static resource allocations. Specifically, three critical issues arise: 1) the spatial mobility of vehicles causes pronounced spatiotemporal variability in edge service demands; 2) the fixed nature of edge services limits the elasticity of service provisioning; and 3) persistent risks of node failures, stemming from hardware faults or software errors, jeopardize service continuity. To address these challenges, this study proposes a fault-tolerant service placement model that jointly optimizes latency minimization and load balancing. We introduce an elastic service placement framework leveraging deep reinforcement learning (DRL). Simulation results demonstrate that our approach outperforms baseline algorithms, achieving robust single-point fault tolerance while maintaining optimal latency and load balancing.