Joint Deployment and Migration of Service Function Chains for Mobility-Aware Services in an Edge-Cloud Environment

With the proliferation of mobile devices, the demand for high-quality mobile services characterized by low latency and high continuity continues to rise. Ensuring the quality of these services is essential to providing intelligent interconnections among individuals, machines, and devices. In the current mobile network landscape, the adoption of network function virtualization (NFV)-based edge-cloud paradigm places network resources in proximity to end-clients, resulting in a substantial reduction in service delays and enhanced efficiency of mobile service management. Nevertheless, the widely decentralized nature of the edge-cloud environment, combined with the stringent quality of service (QoS) demands of delay-sensitive services in dynamic mobile situations, presents a formidable challenge to the deployment of service function chains (SFC). This paper explores a joint deployment and migration of SFCs for mobility-aware services within an NFV-based edge-cloud environment. We formulate a multistage decision-making problem to tackle dynamic mobility patterns of mobile services, aiming to minimize long-term deployment and migration costs, average service latency, while satisfying diverse QoS constraints and adhering to the physical resource constraints of the edge-cloud environment. To tackle the multistage dynamic SFC deployment challenge outlined above, we introduce a deep reinforcement learning (DRL)-based online algorithm. This algorithm autonomously detects variations in the extensively distributed and intricate edge-cloud environment. It generates online deployment solutions without human intervention, facilitating adaptive and rapid service provisioning. Additionally, the algorithm supports the migration of mobile services, ensuring the QoS for high-mobility service requests with low-latency characteristics. Comprehensive experimental results showcase the superiority of our proposed scheme over competing approaches, exhibiting notable advantages in end-to-end latency, resource usage cost, and migration cost. Our scheme achieves average reductions of 5.14%, 3.38%, and 10.67% in these respective metrics, respectively. Furthermore, it enhances the average service acceptance rate by 14.13% under low latency requirements.