DAMP: Dependency-Aware Microservice Placement in Vehicular Edge Computing

Efficient microservice placement in Vehicular Edge Computing (VEC) is critical to ensure low latency, high resource utilization, and service-chain integrity under dynamic vehicular mobility. This paper proposes DAMP, a dependencyaware microservice placement framework implemented using a customized PPO agent (DAMP–PPO) that jointly minimizes endto-end service latency, migration overhead, and cloud offloading while maximizing edge resource utilization. The placement problem is formulated as a constrained Markov Decision Process (MDP) and optimized through a unified policy learning strategy that incorporates dependency awareness, mobility-induced variability, and strict resource feasibility. Action masking and deterministic feasibility repair guarantee valid placements under capacity constraints, while the reward formulation integrates latency, migration, and chain-integrity objectives. The framework is evaluated using real SUMO-generated mobility traces from Luxembourg City across five representative scenarios: Balanced, Stress-CPU, Stress-Bandwidth, Stress-Load, and ForceSplit-Edge. Experimental results show that DAMP–PPO significantly reduces average and 95th-percentile service latency, lowers dependency delay and migration overhead, and improves edge resource efficiency compared with Random, Greedy, A2C, and DQN baselines. These results demonstrate the effectiveness and scalability of the proposed framework for real-time, dependencyaware microservice placement in dynamic VEC environments.