Intent-based Network Management

In today's fast-evolving technological landscape, network management faces the interrelated challenges of increasing network complexity, sophisticated business requirements, and human oversight. Autonomic networks present a promising solution to these challenges. We present a comprehensive management system that integrates intents, a policy-based paradigm, autonomic control loops, large language models, and an assurance framework. Our system utilizes IBN to formalize intents, abstracting complexities from users, and autonomic networking to establish Monitor-Analyze-Plan-Execute (MAPE) loops for intent fulfillment and assurance. By implementing a policy-based approach, we support requirements at varying levels of abstraction and introduce an Application Programming Interface (API) layer to simplify management tasks. We also propose a formal policy information model to ensure consistent mapping and strong inter-policy consistency across abstraction layers. The intelligent decomposition of intents into actionable steps is challenging. Our approach models functional abstractions and decomposes intents into a policy hierarchy, utilizing closed control loop automation guided by Finite State Machines (FSM) to execute policies and deploy intents. We align our methodology with the Metro Ethernet Forum (MEF) Policy Driven Orchestration (PDO) model to enhance applicability and we were one of the first to explore opportunities for IBN in large language models (LLMs). Specifically, we leveraged the few-shot capability of LLMs to progressively decompose intents and generate required actions using a policy-based abstraction, enabling automated policy execution through a closed control loop. Policies are mapped to APIs, forming application management loops for monitoring, analysis, planning, and execution. The practical realization of an intent-based system involves challenges in processing intents and ensuring conformance amidst dynamic network conditions. Thus, we define an assurance framework to detect and act on intent drift, leveraging AI-driven policies generated by LLMs for fulfillment and assurance. Additionally, we utilize a controller design to manage resources and extend the system capabilities to generalize, ensuring a robust and scalable approach to modern network management.

Department dc:contributor.department

Electrical and Computer Engineering

Year dc:date.issued

2024

Author dc:creator

• Dzeparoska, Kristina

Advisor dc:contributor.advisor

• Leon-Garcia, Alberto A.L.G.

• Artificial Intelligence (AI)
• Autonomic System: Closed Control Loop
• Intent-based System
• Large Language Model (LLM)
• Network Automation
• Policy-based Management

Statement dc:rights

• Attribution-NonCommercial-ShareAlike 4.0 International

Licence dc:rights.uri

• http://creativecommons.org/licenses/by-nc-sa/4.0/

Handle dc:identifier.uri

http://hdl.handle.net/1807/141281

OAI identifier oai:identifier

oai:utoronto.scholaris.ca:1807/141281