Intelligent Network System: Service Provisioning Using AI-Agents
Résumé
The network slicing and artificial intelligence (AI) technologies will be the key enablers in the next-generation networks. The substrate network is characterized by an edgecore continuum where the network resources are distributed across the edge and core nodes in the physical network. The existing methods of network slice orchestration face significant challenges in managing dynamic network topology and virtual network embedding (VNE) while scaling efficiently and meeting quality of service (QoS) demands across this continuum. In this paper we have developed an intelligent network system, where the heterogeneous services are provisioned using AI agents. We use the deep reinforcement learning (DRL) agent for admission control and the Large Language Model (LLM) Retrieval-Augmented Generation (RAG) Agents for VNE and network topology management. LLMs offers real-time adaptation and decision-making in dynamic network system. The LLM obtains the topology information using the natural language understanding applied to the retrieved configuration files that may be defined in various formats depending on the motioning tool. The LLM generates the VNE policy and seamless integrates across the edge-core continuum and substrate networks. By leveraging LLMs, we get network automation, adaptability, and predictive capabilities without need of strict formatting of network topology configuration information and service instantiation. Overall, our agentic-AI solution improves the profit (>80%) and successful orchestration (>47%) of heterogeneous services, as compared to the benchmarks.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |