Deep Reinforcement Learning for Topology-Aware VNF Resource Prediction in NFV Environments

Network Function Virtualisation (NFV) has emerged as a key paradigm in network softwarisation, enabling virtualisation in future generation networks. Once deployed, the Virtual Network Functions (VNFs) in an NFV application's Service Function Chain (SFC) experience dynamic fluctuations in network traffic and requests, which necessitates dynamic scaling of resource instances. Dynamic resource management is a critical challenge in virtualised environments, specifically while balancing the trade-off between efficiency and reliability. Since provisioning of virtual infrastructures is time-consuming, this negates the Quality of Service (QoS) requirements and reliability criterion in latency-critical applications such as autonomous driving. This calls for predictive scaling decisions to balance the provisioning time sink, with a methodology that preserves the topological dependencies between the nodes in an SFC for effective resource forecasting. To address this, we propose the model for an Asynchronous Deep Reinforcement Learning (DRL) enhanced Graph Neural Networks (GNN) for topology-aware VNF resource prediction in dynamic NFV environments.