Assuring Efficient Path Selection in an Intent-Based Networking System: A Graph Neural Networks and Deep Reinforcement Learning Approach

Abstract

The recent advancements in network systems, including Software-Defined Networking (SDN), Network Functions Virtualization (NFV), and cloud networking, have significantly enhanced network management. These technologies increase efficiency, reduce manual efforts, and improve agility in deploying new services. They also enable scalable network resources, facilitate handling demand surges, and provide efficient access to innovative solutions. Despite these advancements, the performance of interconnected nodes is still influenced by the heterogeneity of network infrastructure and the capabilities of physical links. This work introduces a comprehensive solution addressing these challenges through Intent-Based Networking (IBN). Our approach utilizes IBN for defining high-level service requirements (QoS) tailored to individual node specifications. Further, we integrate a Graph Neural Network (GNN) to model the network’s overlay topology and understand the behavior of nodes and links. This integration enables the translation of defined intents into optimal paths between end-to-end nodes, ensuring efficient path selection. Additionally, our system incorporates Deep Deterministic Policy Gradients (DDPG) for dynamic weight calculation of QoS metrics to adjust the link cost assigned to network paths based on performance metrics, ensuring the network adapts to the specified QoS intents. The proposed solution has been implemented as an IBN system design comprising an intent definition manager, a GNN model for optimal path selection, an Off-Platform Application (OPA) for policy creation, an assurance module consisting of the DDPG mechanism, and a real-time monitoring system. This design ensures continuous efficient path selection assurance, dynamically adapting to changing conditions and maintaining optimal service levels per the defined intents.